Article

A decision support system for detecting products missing from the shelf based on heuristic rules

Authors:
To read the full-text of this research, you can request a copy directly from the authors.

Abstract

The problem of products missing from the shelf is a major one in the grocery retail sector, as it leads to lost sales and decreased consumer loyalty. Yet, the possibilities for detecting and measuring an “out-of-shelf” situation are limited. In this paper we suggest the employment of machine-learning techniques in order to develop a rule-based Decision Support System for automatically detecting products that are not on the shelf based on sales and other data. Results up-to-now suggest that rules related with the detection of “out-of-shelf” products are characterized by acceptable levels of predictive accuracy and problem coverage.

No full-text available

Request Full-text Paper PDF

To read the full-text of this research,
you can request a copy directly from the authors.

... Prior studies have provided solutions for complex decision problems by applying decision support techniques. A rational and effective approach is to evaluate the decision alternatives as an MADM problem [6][7][8][9][10][11][12][13]. Specifically, to address the task assignment issues, we apply if-then decision rules [7,8,14,15], Techniques for Order Preference by Similarity to Ideal Solution (TOPSIS) [16], and Simple Additive Weighting (SAW) [10,17] to the problem. ...
... A rational and effective approach is to evaluate the decision alternatives as an MADM problem [6][7][8][9][10][11][12][13]. Specifically, to address the task assignment issues, we apply if-then decision rules [7,8,14,15], Techniques for Order Preference by Similarity to Ideal Solution (TOPSIS) [16], and Simple Additive Weighting (SAW) [10,17] to the problem. The logic of TOPSIS is to identify an alternative which has the shortest distance from the positive-ideal solution, such as profitability, and has the longest distance from negative-ideal solution, such as cost. ...
... Then, the preference values obtained using Equations (6) to (8) and shown in Table XI suggest that A5 is the best analyst for T1. For the multiple tasks assignment scenario, we assign the 10 tasks together. ...
... A natural way for manufacturers to detect and predict OOS events without physical audits is to detect anomalies in the transactional data provided by retailers. It has recently been observed that such a process can be automated with a Machine Learning (ML) algorithm [7][8][9]. This approach was first pioneered by [8]. ...
... It has recently been observed that such a process can be automated with a Machine Learning (ML) algorithm [7][8][9]. This approach was first pioneered by [8]. In [9], inventory control models and classification methods were studied. ...
... The methodology that we propose differs from previous approaches in the academic literature [7][8][9][10][11] in several ways. Previous approaches all use sales data to detect OOS events. ...
Article
Full-text available
For decades, Out-of-Stock (OOS) events have been a problem for retailers and manufacturers. In grocery retailing, an OOS event is used to characterize the condition in which customers do not find a certain commodity while attempting to buy it. This paper focuses on addressing this problem from a manufacturer’s perspective, conducting a case study in a retail packaged foods manufacturing company located in Latin America. We developed two machine learning based systems to detect OOS events automatically. The first is based on a single Random Forest classifier with balanced data, and the second is an ensemble of six different classification algorithms. We used transactional data from the manufacturer information system and physical audits. The novelty of this work is our use of new predictor variables of OOS events. The system was successfully implemented and tested in a retail packaged foods manufacturer company. By incorporating the new predictive variables in our Random Forest and Ensemble classifier, we were able to improve their system’s predictive power. In particular, the Random Forest classifier presented the best performance in a real-world setting, achieving a detection precision of 72% and identifying 68% of the total OOS events. Finally, the incorporation of our new predictor variables allowed us to improve the performance of the Random Forest by 0.24 points in the F-measure.
... Ironically, there are also missed sales opportunities when products are not truly out-of-stock but only out-of-shelf. A product is defined as out-of-shelf when it exists inside the store, thus not out-of-stock, but it cannot be found by a customer willing to purchase it [38]. The out-of-shelf condition has two main causes: (i) replenishment errors, i.e. the product is available in the backroom and not in the sales floor area, and (ii) placement errors, i.e. the product is misplaced somewhere in the sales floor area and it is not where it ought to be [42]. ...
... Products being out of stock is a pressing problem causing missed sales opportunities. Several researchers have addressed detection of out-of-stock situations [38,34,37]. Papakiriakopoulos et al. [38] rely on a heuristic rule based approach to detect missing products, as at that time they judged RFID to be not yet operational for this purpose. ...
... Several researchers have addressed detection of out-of-stock situations [38,34,37]. Papakiriakopoulos et al. [38] rely on a heuristic rule based approach to detect missing products, as at that time they judged RFID to be not yet operational for this purpose. Li et al. [34] focus on improving the detection rate of RFID systems on a technical sensor level to identify missing products and distinguish those from products that are there, but hidden to the system through the problem of tag collisions during read. ...
Article
Accurate and timely provisioning of products to the customers is essential in retail environments to avoid missed sales opportunities. One cause for missed sales is that products are misplaced in the store. This can be addressed by fast and accurately detecting those misplacements. A problem of current detection methods for misplaced products is their reliance on up-to-date planogram information, which is often missing in practice. This paper investigates the effectiveness and efficiency of outlier detection methods for finding misplaced products without planograms. To that end, we conduct simulation studies with realistic parameters for different store parameters and sensor infrastructure settings. We also evaluate the detection methods in a real setting with an RFID inventory robot. The findings indicate that our proposed MiProD aggregation of individual detection methods consistently outperforms individual techniques in detecting misplaced products.
... The problem of products missing from the shelf (or "out-of-shelf" problem) is still a frequent phenomenon in the grocery retail sector and can lead to lost sales and decreases consumer loyalty [1]. The term "out-of-shelf" (OOS) is used to describe situation where a customer does not find the product (or sufficient number of that product) one wishes to buy on the shelf of a supermar- ket. ...
... Recently, some published papers suggested solutions of OOS problem through machine-learning technique [1] or classification methods [2,3] in order to automati- cally identify products missing from the shelf. Apart from these approaches, there are some other technolo- gies, which can provide product availability monitoring on retail shelves such as: radio frequency identification (RFID) [4,5,6], computer vision has developed power- ful algorithms for pattern recognition [7], infrared sen- sors [8], weight-sensitive foam [9], etc. ...
Article
Full-text available
Products availability on the shelf, in the store, is a very important aspect in order to provide consumer satisfaction. Instead of conducting physical store audit, this paper demonstrates a counter of the exact number of the products on the shelf prototype realization. The complete test platform on Plexiglas has been developed. The proposed principle is based on interdigitated capacitor fabricated using ink-jet technology, where one set of silver electrodes is posted on the platform and another set on the boxes (products). When all products are on the test platform (or shelf ) the capacitance has a maximum value. Taking products from the shelf this capacitance decreases and complete hardware solution has been made to transfer measured capacitance into number of products and to show this number on the display and turn on warning messages, if it is necessary. The proposed solution enables early detection of low number of products on the shelf and timely replenishment. This paper studies performance of interdigitated capacitors printed with ink-jet printing technology on flexible substrate. Focus point of research presented in this paper is influences of different factors (number of capacitors connected in parallel, accurate position of electrodes of one capacitor, frequency range, etc.) on the capacitance of the structure. Main results are in the demonstration of capacitance change due to the influence of position between electrodes. Test platform, using this principle, has been made to demonstrate that it can be used as an effective solution for out-of-shelf problem.
... In the last decade, several strategies for automatic OOS detection in retail environments have been proposed. Some approaches were quite simple and employed physical sensors or analytical observations [7][12] [13]. More recently, image processing techniques have been employed [11][19] to achieve better performance. ...
... This allowed to monitor the availability of on-shelf products and hence refill OOS. Papakiriakopoulos et al. [13] proposed a machine learningbased approach which used data available in the store information system to check the availability of products. ...
... This generates a problem in demand forecasting, which should be the starting point for all operations planning, and plays a key role in supporting the achievement of company's strategic targets (Moon, Mentzer, Smith, & Garver, 1998). Several literature examples, referring to shelf out-of-stock events, mainly focus on illustrating the consumers' reactions and behavior (Campo, Gijsbrechts, & Nisol, 2000; Emmelhainz, Emmelhainz, & Stock, 1991; Papakiriakopoulos et al. 2008). Indeed, customer satisfaction is a key parameter to increase consumer loyalty towards the brand, specifically in fashion and apparel industry: Campo, Gijsbrechts, & Nisol (2003) estimated the costs incurred by the retailer and the supplier according to the various reactions that consumers may have when facing an out-of-stock situation. ...
... The expression shelf out-of-stock describes the situation where a consumer cannot buy the desired product from stores shelves because it is sold out. The major variables that can affect product availability in the stores and that can be the cause of out-of-stock have been pointed out by several authors in literature (see Papakiriakopoulos, Pramatari, & Doukidis, 2008): ...
Article
Full-text available
Failures occurring in each logistic chain node inevitably affect products availability in storage and distribution points, leading to stock-outs and subsequent customer dissatisfaction. Dealing with retailers which sell to final consumers, the economic estimation of the Shelf Out-of-Stock (OOS) loss is notoriously challenging. Moreover, in fashion and apparel stores, it is even difficult to estimate the size of OOS: due to the fickleness of the shopper, a OOS condition may even not lead to a lost sale. This paper focuses on the verification of the occurrence of out-of-stock events in fashion stores, aiming to get a quantitative evaluation of the potential lost sales through the analysis of the number of days of products unavailability. The number of OOS events due to early stock depletion will be consequently calculated, along with their consequences. The proposed procedure has been validated on real data of an important Italian fashion company.
... Product availability in retail stores is often described and analysed through out-ofstock problem (Ettouzani et al, 2012), where the OOS rate was also most frequently used as its basic indicator. Attention has been devoted to its measuring (Roland Berger Consultants, 2003;Gruen, Corsten, 2007), identifying (Papakiriakopoulos et al., 2009;Papakiriakopoulos, Doukidis, 2011) main root causes (Fernie, Grant, 2008;Ehrenthal, Stolzle, 2013), effects (Gruen, 2007;Musalem et al., 2010), and customer responses in out-of-stock situations (van Woensel et al., 2007;Zinn, Liu, 2008). ...
... Under the auspices of ECR Europe, by analysing the data on daily sales, Hausruckinger (2005) set the principles for calculating the boundaries within which the expected sales ranges. However, as his approach can only be applied to a small number of FMCG products with low sales volatility (Papakiriakopoulos et al., 2009), for calculations in this study we also relied on features proposed by Papakiriakopoulos, Doukidis (2011). ...
Article
Full-text available
By increasing inventories, retailers attempt to raise service levels, and thus increase sale. However, in addition to a positive impact on product availability and sale, higher inventory levels may cause problems in performing in-store activities. As poor backroom-to-shelf replenishment process emerges as one of the most common causes of stock-out situations, this article compares store and on-shelf FMCG product availability at SKU level in different stores of a single retailer. In relation to this, besides direct, we have also investigated the indirect effect of inventory level on sale, by using store and shelf out-of-stocks as mediators. The results of the research showing much higher level of shelf- compared to store stock-out rate confirmed the existence of the problem in the realization of internal product flows within retail stores. However, despite the occurrence of this problem, besides direct positive effect of inventory level on sale, its indirect effect was positive aswell. Therefore, these results were analysed in the context of other similar studies. In addition to empirical research, the article also discusses certain implications of more efficient organisation of in-store activities.
... Analytical data-driven techniques methodologically rely on statistical learning (Hastie, Tibshirani, & Friedman, 2009), that is, the partitioning of large amounts of empirical observations into training and testing sets to evaluate the models and conclusions induced from observations. The current work contributes to growing literature that uses statistical learning techniques to generate shopper insights (Cil, 2012;Oestreicher-Singer, Libai, Sivan, Carmi, & Yassin, 2013;Papakiriakopoulos, Pramatari, & Doukidis, 2009) while also responding to recent calls to identify new phenomena of interest for scholarly marketing research. ...
Article
Shoppers enter stores to meet diverse need states, such as getting a soft drink and a bag of potato chips for immediate consumption or purchasing ingredients for meals to make on the same day. The present article introduces the concept of a shopping mission that can be identified by the complementarity and sales affinity among the balanced number of product categories in trips with relatively concrete goals. The authors develop an analytical method to identify several shopping missions at the store level and demonstrate the utility, validity, and replicability of this method using a data set with 4 million baskets from a multinational supermarket chain. The authors also provide evidence that the identification of shopping missions offers a wealth of opportunities for targeted shopper marketing activities.
... A general consequence of a shrinkage event is the loss or disappearance of items. While the source of shrinkage can take many forms such as damage, misplacement (e.g., [17]), process error, shop-lifting (e.g., [7]), spoilage (e.g., [18]), theft (e.g., [9]) and ticket-switching [25,26], the loss per shrinkage incident depends on the number and type of items involved, the short-term demand for such items and their substitutes, the temporal aspect of that item (e.g., fashion apparel, perishables), and the complexity of the item's replenishment process. Regardless, while a majority of shrinkage incidents affect only one (type of) item (e.g., a pair of jeans), some simultaneously involve multiple items per incident. ...
Article
Inventory inaccuracies in retail stores result from a combination of controllable and uncontrollable factors such as theft, damage, spoilage, misplacement, process errors, ticket-switching, among others. While most shrinkage types affect only one (type of) item, ticket-switching simultaneously affects the inventory of multiple items. Ticket-switching is the process of switching the identifier or ticket of an expensive item with that from a (relatively) cheap item with the explicit intent of purchasing the expensive item by paying the cheap item’s price. Ticket-switching incidents distort inventory records in store information systems. Inventory management decisions based on such data from store information systems are therefore sub-optimal. We study the effects of ticket-switching on optimal order quantity of the involved items and the resulting profit. Under uniformly distributed demand and yield conditions, we find that ticket-switching increases (decreases) the optimal order quantity of the expensive (cheap) items. Surprisingly, results from our analysis indicate that profit on expensive (cheap) items are higher (lower) in the presence of ticket-switching behavior than otherwise.
... The literature has many studies investigating the OOS problem. Some of the researchers focused on identifying OOS situations (Papakiriakopoulos & Doukidis, 2011;Papakiriakopoulos et al., 2009), measuring OOS (Corsten & Gruen, 2005), understanding the effects of OOS (Gruen & Corsten, 2007;Musalem et al., 2010), analyzing the main root causes (Ehrenthal & Stolzle, 2013;Fernie & Grant, 2008) and investigating customer responses in OOS situations (Zinn & Liu, 2008;Van Woensel et al., 2007). ...
Article
Full-text available
The purpose of this study is to predict the share of visual inventory (SOVI), which is defined as the number of stock-keeping units (SKUs) of a company’s products, calculated as a percentage of the total SKUs on the display of all products. Research studies in the past have focused mainly on the impact of inventory, which includes back end and visual inventory, on sales but less attention has been given to the impact of SOVI on sales. To address this research gap, this study attempted to create an analytics model to predict SOVI at the category of soft drinks level using four predictor variables namely point of purchase display, channel/sub-channel, package group, product category, and derived variable gross national income (GNI). The results were encouraging confirming the effectiveness of such a model. The researchers utilized a data set collected over a period of 18 months (February 2016 to July 2017) by a soft drink firm headquartered in the United States. Based on the findings, it is suggested that this prediction model can be utilized by other researchers and practitioners to predict SOVI of other soft drinks, fast-moving consumer goods (FMCG), and food and beverage companies.
... One of the problems in smart packaging is so called out-of-shelf (OOS) problem. This problem of products missing from the shelf is still a frequent phenomenon in the grocery retail sector and can lead to lost sales and decreases consumer loyalty [12]. The term "out-of-shelf" (OOS) is used to describe situation where a consumer does not find the product (or sufficient number of that product) one wishes to buy on the shelf of a supermarket, during a shopping tour. ...
Conference Paper
Full-text available
Printed electronics on flexible substrates is new approach to electronics in last few years. This paper studies performance of interdigitated capacitors printed with ink-jet printing technology on flexible substrate. Focus point of research presented in this paper is influences of different factors (number of capacitors connected in parallel, number of “fingers” of single capacitor, accurate position of electrodes of one capacitor, etc.) on the capacitance of the structure. Main results are in the demonstration of capacitance change due to the influence of position between electrodes. Test platform, using this principle, has been made to demonstrate that it can be used as an effective solution for out-of-shelf problem.
... (1) ECR Europe presented the approach for calculating the expected sales ranges for each product (Hausruckinger, 2006). However, bearing in mind that this method is problematic for FMCG products with high sales volatility (Papakiriakopoulos, Pramatari & Doukidis, 2009), following Grubor & Milicevic (2015), when computing variables in formula (1), we used additional features proposed by Papakiriakopoulos and Doukidis (2011). All necessary data from 2013 (2014), were obtained from retailers' ERP information platforms connected with stores' POS terminals. ...
Article
Full-text available
Bearing in mind that it represents one of the main preconditions of sales, product availability is the key task of retail companies and their delivery systems. This paper analyses its levels from the aspect of centralized and DSD systems. The research is conducted in Serbia, Bosnia and Herzegovina and Montenegro, including 84 stores and more than 70 FMCG products per each store. Thereby, the comparisons in product availability levels between alternative delivery systems are carried out within different trading formats and within different product categories. Unlike the results of similar studies and ongoing changes on retail markets, this research shows that at retailers in these Western Balkan countries, availability levels are higher in the case of DSD system.
... Inventory shrinkage, a significant issue in retail stores, arises due to several sources including employee-theft [18], shoplifting [5], process errors, vendor errors, item misplacement [15], spoilage of perishables [16], breakage during handling, ticket-switching [27], among others. Among these, employee-theft and shoplifting are generally considered to be major factors that add to about 80% of the overall retail store inventory shrinkage [4]. ...
Article
Ticket-switching incidents simultaneously and directly affect the actual and store information system inventory of multiple products. We model the discrepancy in inventory information stored in the retailer's information system vs. reality and related consequences for the customer and the retail store. We also consider a retail store with constrained shelf-space availability and study how this retailer should optimally allocate shelf-space between these products when ticket-switching is present. We model this by taking into account the customer arrival sequence. For customers who use online store inventory information, our results indicate that in the presence of ticket-switching, an item can be guaranteed to be in-stock at the store for immediate pick-up only for ‘cheap’ items. The results from this study also have policy implications for retail stores in terms of computing benefit estimates as well as shelf-space allocation.
... Indeed, previous work has successfully shown the competitive accuracy performance of associative classifiers, such as CMAR [16], CPAR [28], HARMONY [24], LAC [23] and GARC [5]. Associative classification has been popularly applied in practice, such as detecting products missing from the shelf [19], maximizing customer satisfaction [14] and predicting patients' hospitalization [27]. However, in the associative classification rule generating process, a large number of rules can be generated, and, thus, there can be many redundant or irrelevant rules. ...
Article
Associative classifiers have been proposed to achieve an accurate model with each individual rule being interpretable. However, existing associative classifiers often consist of a large number of rules and, thus, can be difficult to interpret. We show that associative classifiers consisting of an ordered rule set can be represented as a tree model. From this view, it is clear that these classifiers are restricted in that at least one child node of a non-leaf node is never split. We propose a new tree model, i.e., condition-based tree (CBT), to relax the restriction. Furthermore, we also propose an algorithm to transform a CBT to an ordered rule set with concise rule conditions. This ordered rule set is referred to as a condition-based classifier (CBC). Thus, the interpretability of an associative classifier is maintained, but more expressive models are possible. The rule transformation algorithm can be also applied to regular binary decision trees to extract an ordered set of rules with simple rule conditions. Feature selection is applied to a binary representation of conditions to simplify/improve the models further. Experimental studies show that CBC has competitive accuracy performance, and has a significantly smaller number of rules (median of 10 rules per data set) than well-known associative classifiers such as CBA (median of 47) and GARC (median of 21). CBC with feature selection has even a smaller number of rules.
... For example, it might be interesting to analyze the extent to which more sophisticated policies can compete with RFID (e.g., [22]). Finally, we see great potential in the combination of RFID with machinelearning techniques for the detection of stock-outs (e.g., [12,30]). ...
Article
This contribution is concerned with the value of RFID for retail store operations, particularly the use of the technology to automate shelf replenishment decisions. We construct and test an inventory control policy based on RFID data with case-level tagging. Our model incorporates RFID hardware capable of detecting bidirectional product movements between a store's backroom and the sales floor. In contrast to prior research, we account for detection errors caused by imperfect RFID read rates. Furthermore, we propose and evaluate a simple heuristic extension to avoid some of the inherent downsides of fully automatic inventory control. We compare the performance of these policies under stochastic demand, lost sales, and shrinkage to the traditional scheme with periodic reviews in a simulation study. Our results indicate that RFID-based policies have the potential to improve cost efficiency and service levels. However, different sensitivities to cost factors and suboptimal read rates must be considered when choosing a policy.
... RFID technology has been extensively studied as a means of tracking materials in supply chains for example in retailing (e.g., Papakiriakopoulos et al., 2009; Thiesse et al., 2009; Subramanian and Iyigungor, 2006; Prater et al., 2005) and in manufacturing industry (Irani et al., 2010; Ngai et al., 2008b; Chao et al., 2007). Applications in service businesses have attracted fewer researchers' attention (Ferrer et al., 2010; Slettemeås, 2009; Smith et al., 2009). ...
Article
Full-text available
This study focuses on the potential for value creation in combining radio frequency identification (RFID) technology with social media marketing. We analyse a value network comprising an RFID technology provider, a software supplier, a sports centre adopting the system, and the sports centre's end customers. The findings suggest that value creation related to new technology requires deep collaboration between the suppliers, as they need to develop value propositions that motivate companies to adopt new systems and end customers to use these systems. The threat to the end customer of losing privacy when RFID technology is employed, and the interactive nature of communication in social media, means that the value created by a system combining RFID technology and social media is highly dependent on the end customer's motivation to be involved in value creation.
... The out-of-stock problem in retail trade has been analysed from various aspects. Attention has been devoted to measuring (Roland Berger Consultants, 2003;Gruen & Corsten, 2007), identifying (Papakiriakopoulos et al., 2008;Papakiriakopoulos & Doukidis, 2011), main root causes (McKinnon et al., 2007;Fernie & Grant, 2008), effects (Anderson et al., 2006;Gruen, 2007;Musalem et al., 2010), and specifics of customer responses in OOS situations (Sloot et al., 2005;van Woensel et al., 2007;Zinn & Liu, 2008). Besides brick-and-mortar conditions, this problem has been analyzed in an online context too (Breugelmans et al., 2006;Jing & Lewis, 2011;Pizzi & Scarpi, 2013). ...
Article
Full-text available
Seeking to reduce consumer purchasing effort, retailers strive to provide adequate product availability levels in retail stores. Otherwise, consumers may find themselves in stock-out situation, which may be a waste of their time, money and energy. As the mentioned situation intends to make a negative impact on the business performance of retailers and their suppliers, the issue of on-shelf product availability deserves particular attention. This article analyses this issue from the aspect of certain characteristics of product sales. Product availability, expressed by means of out-of-stock rate, is compared within various categories of sales variation and speed of turnover variables. In addition to these characteristics, product availability is also analysed from the aspect of various retail formats. The results have indicated that products with the highest sales variation, as well as slow turning products, are the most problematic in terms of on-shelf availability; in other words, they are characterised by the highest out-of-stock (OOS) rates. On the other hand, unlike large-scale retail formats, average store availability is the lowest in superettes, which indicates that the average OOS rate declines with the size of retail formats. In addition to serving as a basis for further research related to product availability in retailing, the given results may be used by retail managers when optimising customer service levels. © 2014, Kauno Technologijos Universitetas. All rights reserved.
... Such figures are, however, conservative due to the inherent inaccuracy of current OOS measurement methods and the increased difficulty to track shelf OOS. Shelf OOS cannot be detected by perpetual inventory (PI) systems and rely primarily on visual checks or some algorithmic approach (Papakiriakopoulos et al., 2009). If store personnel or an OOS-alerting algorithm do not notice a shelf OOS, that OOS will go undetected and will not be incorporated into the OOS rate estimations. ...
Article
Full-text available
Purpose – The purpose of this paper is to: first, provide a systematic review of the drivers of retail on-shelf availability (OSA) that have been scrutinized in the literature; second, identify areas where further scrutiny is needed; and third, critically reflect on current conceptualizations of OSA and suggest alternative perspectives that may help guide future investigations. Design/methodology/approach – A systematic approach is adopted wherein nine leading journals in logistics, supply chain management, operations management, and retailing are systematically scanned for articles discussing OSA drivers. The respective journals’ websites are used as the primary platform for scanning, with Google Scholar serving as a secondary platform for completeness. Journal articles are carefully read and their respective relevance assessed. A final set of 73 articles is retained and thoroughly reviewed for the purpose of this research. The systematic nature of the review minimizes researcher bias, ensures reasonable completeness, maximizes reliability, and enables replicability. Findings – Five categories of drivers of OSA are identified. The first four – i.e., operational, behavioral, managerial, and coordination drivers – stem from failures at the planning or execution stages of retail operations. The fifth category – systemic drivers – encompasses contingency factors that amplify the effect of supply chain failures on OSA. The review also indicates that most non-systemic OOS could be traced back to incentive misalignments within and across supply chain partners. Originality/value – This research consolidates past findings on the drivers of OSA and provides valuable insights as to areas where further research may be needed. It also offers forward-looking perspectives that could help advance research on the drivers of OSA. For example, the authors invite the research community to revisit the pervasive underlying assumption that OSA is an absolute imperative and question the unidirectional relationship that higher OSA is necessarily better. The authors initiate an open dialogue to approach OSA as a service-level parameter, rather than a maximizable outcome, as indicated by inventory theory.
... Mou et al. (2018) reviewed 255 papers on retail store sales, some of which used POS data. POS data have also been used for estimating shelf inventory level (Papakiriakopoulos et al. 2009) to avoid shelf stock-outs. ...
Article
Full-text available
Many studies examined negative causal effects between items sold in retail stores, including sales cannibalization, a process by which the sales of new products and sales promotions possibly decrease the sales of other products. Souvenir shops usually have only one opportunity to sell products to each visitor because of their nature. Thus, clarifying the contemporaneous causal relationships among the sales of products sold in souvenir shops is much more important. We propose that there are informative causal relations among the sales of tourism products. We develop a methodology to estimate these causalities given exogenous factors without specifying the causal directions. The proposed model is a structural vector auto regressive model with exogenous variables, and its estimation method is a modified vector auto regressive model to a linear non-Gaussian cyclic model. Applying the proposed methodology, we analyzed point of sale data from a souvenir shop. The results show that tourists who purchase tourism products of an ideal size for office/workplace souvenirs tend to simultaneously purchase other tourism products of the same size or products for households. The obtained result can help marketers plan sales promotions and inventory controls.
... Im Folgenden wird unter einer Out-of-Shelf-Situation ein Zustand in einer Filiale verstanden, bei dem Kunden einen bestimmten Artikel nicht mehr am Regalplatz vorfinden, sei es weil dieser in der gesamten Filiale ausverkauft ist (Out-of-Stock-Situation als Spezialfall der Out-of-Shelf-Situation), oder aber, der nicht im Regal verfügbar ist, obwohl er in der Filiale, beispielsweise im Lager oder am Werbe-Aufbau, verfügbar wäre (Papakiriakopoulos et al. 2008). ...
Chapter
Full-text available
Zwei der zentralen Technologietrends der Digitalisierung sind Big Data und Künstliche Intelligenz (KI), insbesondere Machine Learning. Es wird mitunter in der vor allem von Informatikern und Technologen dominierten Literatur der Eindruck erweckt, dass Technologien für Unternehmen unmittelbar einen Mehrwert stiften. Inwieweit allerdings ein Analogieschluss beispielsweise aus Googles Erfolgen mit AlphaGo bis MuZero auf primär betriebswirtschaftliche Problemstellungen zulässig ist, soll für die Domäne des Handels in dem vorliegenden Kapitel untersucht werden. Aufbauend auf einer grundsätzlichen Erörterung des Big-Data-Phänomens aus einer Entscheidungsperspektive werden Einsatzmöglichkeiten für das Marketing im Handel untersucht. Im letzten Abschnitt wird problematisiert, wie Machine Learning in ausgewählten Bereichen Mehrwerte für Unternehmen eröffnet.
... Lost sales phenomenon is also highlighted by marketing literature from another point of view. Several literature examples, referring to shelf OOS events, mainly focus on illustrating the behaviour of consumers (Papakiriakopoulos et al., 2009). The results show that the customer behaviour facing OOS depends on product, brand and POS. ...
Article
Many retailers experience significant losses due to out of stock (OOS) risk. A thorough analysis of a case study has revealed that this situation depends on three main factors: First, the demand and supply risks, second the responses of customers to on shelves OOS and third the behavior of managers facing these risks. This exploratory and theoretical research focuses on the modern retail supply chain risks associated with the supply side and the demand side and highlights the impact of these risks on the on shelves OOS and the total cost of lost sales (TCLS). While the literature on supply chain risks is growing, there is a lack of realistic support tools for the prediction of the supply risk when the order size and the delivery lead-time are dependent decision variables. The problem presented here, and the way in which it is addressed, is different from similar problems in the literature. We do not merely assume the lead-time to be a random exogenous variable, but we include the impact of the order size decision on the delivery lead-time and use the result to predict the supply risk. We propose a new approach for ordering decision under supply and demand uncertainties. This approach considers a new method to predict the supply risk and proposes a realistic estimation of the TCLS taking into account the responses of customers to OOS. Some numerical examples from a real case study are presented to illustrate the proposed approach. Keywords: Modern retail supply chain; supply risk; demand risk; lost sales; out-of-stock; and consumer behavior.
... Lost sales phenomenon is also highlighted by marketing literature from another point of view. Several literature examples, referring to shelf OOS events, mainly focus on illustrating the behaviour of consumers (Papakiriakopoulos et al., 2009). The results show that the customer behaviour facing OOS depends on product, brand and POS. ...
Article
Many retailers experience significant losses due to out of stock (OOS) risk. A thorough analysis of a case study has revealed that this situation depends on three main factors: first, the demand and supply risks, second the responses of customers to on shelves OOS and third the behaviour of managers facing these risks. This exploratory and theoretical research focuses on the modern retail supply chain risks associated with the supply side and the demand side and highlights the impact of these risks on the on shelves OOS and the total cost of lost sales (TCLS). We propose a new approach for ordering decision under supply and demand uncertainties. This approach considers a new method to predict the supply risk and proposes a realistic estimation of the TCLS taking into account the responses of customers to OOS. Some numerical examples from a real case study are presented to illustrate the proposed approach.
... The final perspective encountered in the literature deals with the problem of detecting and measuring OOS events objectively. Some authors (Papakiriakopoulos et al., 2009;Moussaoui et al., 2016) have made further analysis of how to use KPIs linked to stockouts as a way of measuring such events and how to make detection easier (with or without technological aids). The literature concerning these KPIs also contains no clear definitions or proposals of ways to measure the events, which are included in the methodological proposal given here. ...
Ensuring “On-shelf Availability” (OSA) or avoiding “Out of stock” events (OOS) in a store is a key factor for customer satisfaction. Likewise, logistics rationalization throughout the supply chain is a key factor for business profitability. The main aim of this paper is twofold: First, to design a participative methodology that systematically seeks to reduce OOS events, maintain a store's image and rationalize logistics processes by adopting the Action Research approach; Second, to illustrate the usefulness of this methodology, by implementing it in one of the largest retailers in Spain. The proposal selects and applies a set of specific key performance indicators (KPIs) related to shelf use by redesigning store and logistics processes in a structured, participative way.
... This important task will be carried out using Principal Component Analysis, computational results, guided with previous studies and in-house specific knowledge about the data.Table 1probably exhibits and explains all important features for the out-of-stock problem found in the database (neobična rečenica). We remark that this set of features is in great detail similar to the set of features used in (Papakiriakopoulos et al., 2009) studies on out-of-stock problems. ...
Conference Paper
Improving inventory management is essential to retailer profitability. This paper proposes a supervised learning approach for Out-of-Stock (OOS) detection by Texture, Color and Geometry features in high-resolution panoramic images of grocery retail shelves. Cascade classifiers are used to detect labels that can potentially be used to confirm the presence of the OOS cases. The image acquisition setup includes a camera cart that shoots from multi-viewpoints aiming a parallel motion to the shelf. The correction of perspective distortion is applied to handle the different camera translation motions while stitching together images with a high-level of similarity. From the generated panoramas, the proposed OOS detection is followed by classification with Support Vector Machines. The experimental tests were performed throughout the retail environment with real data obtained from supermarket shelves containing labels near the visible ruptures. Results show a detection accuracy of 84.5% for OOS and a sensitivity of 86.6% for label detection.
Chapter
In large-scale retail trade, a very significant problem consists in analyzing the response of clients to product promotions. The aim of the project described in this work is the extraction of forecasting models able to estimate the volume of sales involving a product under promotion, together with a prediction of the risk of out of stock events, in which case the sales forecast should be considered potentially underestimated. Our approach consists in developing a multi-class classifier with ordinal classes (lower classes represent smaller numbers of items sold) as opposed to more traditional approaches that translate the problem to a binary-class classification. In order to do that, a proper discretization of sales values is studied, and ad hoc quality measures are provided in order to evaluate the accuracy of forecast models taking into consideration the order of classes. Finally, an overall system for end users is sketched, where the forecasting functionalities are organized in an integrated dashboard.
Article
Purpose – The objective of this paper is to demonstrate the challenges when developing a common performance measurement system (PMS) in the context of a collaborative supply chain. Design/methodology/approach – The paper utilizes qualitative and quantitative data from a case study. The qualitative data refer to the assessment of collaborative performance measures based on interviews with experts, while the quantitative data demonstrate the use of two performance measures in a collaborative supply chain network. Findings – The development of a collaborative PMS is a challenging task. Through the systematic study of two significant performance measures for a supply chain, it was found that the one could not be supported due to reliability restrictions, while the other requires the development of a complex information system. Based on these, a discussion of specific challenges follows. Research limitations/implications – The paper has the general case study limitations. Practical implications – Companies operating in supply chain networks need to synchronize existing business processes and data before the design of a new PMS. Selecting the measures and the measurement method is not a trivial task. Important challenges reveal when dealing with, underlying data, business processes and the evaluation method of a PMS in supply chains. Originality/value – The management control function usually focuses on the design and development of PMSs for a single organization. Limited knowledge exists when more than two companies require the development of a PMS for a jointly agreed business process.
The food industry is under pressure of managing food safety 'from farm to fork'. Food traceability has become an essential task for the food industry in supplying food products with adequate quality and safety assurance. However, it could be a daunting task if the effort on traceability systems is separated from profitable strategies of managing supply chain operations, as traceability will add extra costs for producers and for the food sector as a whole. This paper aims to develop an effective strategy for the integration of traceability initiatives with supply chain operations and create competitive advantages. Two industrial cases are presented to illustrate the proposed applications. It provides the insight into the value of food traceability and its relationship with supply chain management and helps food organisations choose an appropriate traceability solution.
Purpose The purpose of this paper is to present a methodology to structure the problem of retail out of stock (OOS). This methodology allows investigating risk factors and barriers related to the main causes and consequences that lead to OOS occurring. Design/methodology/approach The proposed methodology to structure the OOS problem is based on the bow- ie tool, which allows better visualisation, understanding and analysis of a complete OOS scenario. This proposal comprises exploring the main causes and consequences of OOS, the barriers to prevention and mitigation, the escalation factors to control undesirable events and to define actions to eliminate or mitigate the OSS risk. Findings Several potential causal risk factors, related to technical, behavioural, cultural and organisational aspects, were identified with this methodology. With the analysis of the OOS scenario, it was observed that the factors that lead to the OOS risk are preventable. In order to improve existing barriers or implement new barriers, a set of actions can be recommended to reduce or eliminate OOS risk factors. Originality/value From a better understanding of hazard, the bow-tie methodology allows identifying crucial factors that could be acted upon to reduce the incidence of OSS. Thus, the value is to propose a methodology that allows establishing the preventive and protective barriers necessaries and the escalation factors related to each of these to help structure the problem and consequently reduce the OOS in retail organisations.
Article
When a retail store places an item for sale at a set price, the expectation is that the customer pays this price for the item. However, the ‘customer’ may not necessarily pay this amount due to any number of legitimate (e.g., price promotion) as well as illegitimate (e.g., theft) reasons. We consider ticket-switching, a scenario whereby the customer pays a lower amount for the purchased item by switching its price identifier. We propose the use of item-level RFID tags to address ticket-switching in apparel retail stores. We then develop authentication protocols that are directed at reducing the occurrence of ticket-switching incidents as well as identifying them when they occur. We evaluate the security properties of the proposed protocols.
Chapter
Out-of-stock (OOS) problem is a significant reason of the decline of goods sales in offline supermarkets since the frequent lack of goods on the shelves can reduce the enthusiasm of shoppers. For this purpose, it is necessary to effectively detect the OOS situation, which can ensure that the products are replenished in time. In this paper, an out-of-stock detection method based on deep learning is proposed. We first introduce the Faster R-CNN algorithm to obtain location information. Then the Faster R-CNN algorithm is followed by three out-of-stock detection methods: canny operator, gray level co-occurrence matrix, and color features for out-of-stock detection. The experimental results show that the method based on canny operator performs better, which can achieve a recall rate of 83.9%, a precision rate of 91.7% and a F1-Measure of 87.6% in real scene dataset.
Purpose ‐ The purpose of this paper is to increase our understanding of the causes for stockouts in retailing. Design/methodology/approach ‐ Mixed methods study, using instore observations, interviews with key informants in consumer goods and retailing, and a field study of stockouts and their causes in multiple wholesale stores over two years. Findings ‐ The results indicate that the causes for stockouts are specific to retailer, store, category and item. Improvements to store operations and the coordination of store delivery and shelf replenishment are most effective in reducing stockouts. Manual audits of stockouts and their causes benefit instore execution and provide the level of detail necessary for management to prioritize areas of improvement. Research limitations/implications ‐ Future research may investigate the operational and cost impact of incorporating demand seasonality in shelf replenishment that may lead to an improved coordination of replenishment and demand cycles. Practical implications ‐ A procedure is proposed to help store managers reduce stockouts well below the global average of 8.3 percent. Originality/value ‐ The paper extends the literature by providing a comprehensive set of itemized causes of retail stockouts and reflects implications for sales-data driven research. It adds to the emergent research that applies service-dominant logic to retail stockout research.
Article
Full-text available
This paper proposes an effective decision support system based on an absorbing Markov model, which is used for helping decision makers in Faculty of Computing and Information Technology (FCIT) at King Abdul Aziz University (KAU) in controlling student’s flow transition enrollment. Several important controlling criteria that govern student’s flow performance during semesters are evaluated. These include estimating students flow between different study levels, the average life time a student spends at each level, the semesters required for graduation, and students graduating probability. A complete performance evaluation comparison between boys and girls at IT College is investigated. Results show that girls achieved better performance than boys. The system has several advantages, such as, helping to find any bottle necks to be solved during student’s transition study from one semester to another, and helping to know students needed facilities to planning for future required resources, hence achieving good quality and efficient university education. The proposed model is validated using cross-validation methodology, and the achieved results were acceptable
Article
Aligning staff with changing customer and store needs is key to store operations management. There is considerable research on personnel scheduling, but little on the common phenomenon of Real-Time Labour Allocation (RTLA), where mismatches between workforce supply and demand are addressed by allocating potentially cross-trained employees in real time. Our interviews with retail practitioners confirm that RTLA decisions lack analytical justification. In view of this, we design a generalisable stylised Retail Store Simulator (RSS) and instantiate the RSS using data from a gourmet supermarket. Simulation results show substantial long-term benefits to store performance from RTLA – a potential 6.6% increase in market share compared with No-RTLA. We further discuss RTLA's benefits under various employee cross-training configurations, answering a question from the collaborating retailer – “given the benefits of RTLA, how should we manage workforce flexibility?” We conduct extensive “what-if” analysis and find that broadening employee skill range and deepening employee proficiency increase the benefits of RTLA. This research helps understand workforce management at the execution stage.
Article
Out-of-shelf events refer to periods of time in which items of a certain product are not available to customers. It is clear that incidents of this nature result in economic loss, but their side effects are much more profound: since there is no record of missed sales opportunities, the estimated demand curve tends to be inaccurate. As a result, order placement strategies employed by retailers are based on imprecise forecast models, so further out-of-shelf events are very likely to occur: a vicious cycle, hence, arises. In this work, we propose a multi-agent-based simulation to evaluate the impact of out-of-shelf events that considers the reactions of customers towards these incidents and retailers' ordering strategies. Our results show that these events have a significant effect on demand estimation and that multi-agent-based simulations may provide interesting insights and support for the development of more accurate forecast models in retail.
Article
Shelf out-of-stock (OOS) is a salient problem that causes non-trivial profit loss in retailing. To tackle shelf-OOS that plagues customers, retailers, and suppliers, we develop a decision support model for managers who aim to fix the recurring issue of shelf-OOS through data-driven audits. Specifically, we propose a point-of-sale (POS) data analytics approach and use consecutive zero sales observations in POS data as signals to develop an optimal audit policy. The proposed model considers relevant cost factors, conditional probability of shelf-OOS, and conditional expectation of shelf-OOS duration. We then analyze the impact of relevant cost factors, stochastic transition from non-OOS to OOS, zero sale probability of the underlying demand, managers' perceived OOS likelihood, and even random fixes of shelf-OOS on optimal decisions. We also uncover interesting dynamics between decisions, costs, and probability estimates. After analyzing model behaviors, we perform extensive simulations to validate the economic utility of the proposed data-driven audits, which can be a cost-efficient complement to existing shelf inventory control. We further outline implementation details for the sake of model validation. Particularly, we use Bayesian inference and Markov chain Monte Carlo to develop an estimation framework that ensures all model parameters are empirically grounded. We conclude by articulating practical and theoretical implications of our data-driven audit policy design for retail managers.
Article
Prevalent execution errors such as out-of-stock, inventory record inaccuracy, and product misplacement jeopardize retail performance by causing low on-shelf availability, which discourages not only retailers who have lost sales but also manufacturers who have worked hard to deliver goods into retail stores. Thus, external service companies are hired by manufacturers to conduct manual inspection regularly. Motivated by the practical need of shelf audit service providers, we use a general cost structure to develop a decision support model for periodic inspection. Some qualitative insights about the intricate relationships among inspection efficacy, cost factors, failure rate of shelf inventory integrity, and optimal decisions are derived from analytics assuming risk-neutrality. From simulation experiments we also find that managers' risk preferences have non-trivial impacts on optimal decisions. Based on a total cost standpoint high-quality inspection is predominantly preferred regardless of the level of risk aversion. Finally, we propose a Bayesian statistical model and a Markov chain Monte Carlo approach to estimate model parameters such that managers can make empirically informed decisions. Our major contribution lies in developing a mathematical model that is practically applicable and proposing a Bayesian estimation approach to rationalize unobservable model parameters, which are influential to optimal decisions but often arbitrarily assumed by decision makers.
Article
The out-of-shelf problem is among the most important retail problems. An alerting mechanism that would suggest which products the store manager should consider in order to avoid out-of-shelf instances is considered to be of great value. In trying to encounter this problem, we considered the prevention of out-of-shelf as a classification problem. By utilizing the four different classification methods, we examined the predictive accuracy based on real life data sets. The comparison of the results has been facilitated by Receiver Operating Characteristic Curves and examined both at retail chain and to the store level. Based on the results obtained, we were able to draw brief remarks.
Article
In the digital age, retail store operations face a variety of novel challenges and complexities. In order to assist researchers and practitioners in tackling challenges and exploiting opportunities, we collect 255 articles on retail store operations from 32 operations research, management science, retailing, and general management journals over the period 2008-2016. We review the current state of research within the context of retail store operations based on operations decisions and research themes. By discussing the limitations present in these papers, we identify the existence of a number of research gaps and propose several opportunities for researchers seeking to advance our understanding of operational performance.
Article
Full-text available
The out-of-shelf problem is among the most important retail problems. This work employs two different classification algorithms, C4.5 and naïve Bayes, in order to build a mechanism that makes decisions about whether a product is available on a retail store shelf or not. Following the same classification methods and feature spaces, we examined the classification performance of the algorithms in four different retail chains and utilized ROC curves and the area under curve measure to compare the predictive accuracy. Based on the results obtained for the different retail chains, we identified certain approaches for the development and introduction of such a mechanism in different retail contexts.
Article
Full-text available
Capacity limitations for manufacturers of seasonally demanded goods create difficult problems for both practitioners and researchers. Empirical data suggest that practitioner response is far from optimal. Optimal solutions, however, are precluded for realistic problems due to computational complexity. Here, the structure of optimal policies is explored and heuristics based on myopic policies are developed. For simple problems, the best heuristic deviates from optimality by an average of 2.5% over a variety of conditions. Heuristics are also compared under more realistic business conditions by simulation.
Article
Full-text available
The naive Bayes classifier greatly simplify learn-ing by assuming that features are independent given class. Although independence is generally a poor assumption, in practice naive Bayes often competes well with more sophisticated classifiers. Our broad goal is to understand the data character-istics which affect the performance of naive Bayes. Our approach uses Monte Carlo simulations that al-low a systematic study of classification accuracy for several classes of randomly generated prob-lems. We analyze the impact of the distribution entropy on the classification error, showing that low-entropy feature distributions yield good per-formance of naive Bayes. We also demonstrate that naive Bayes works well for certain nearly-functional feature dependencies, thus reaching its best performance in two opposite cases: completely independent features (as expected) and function-ally dependent features (which is surprising). An-other surprising result is that the accuracy of naive Bayes is not directly correlated with the degree of feature dependencies measured as the class-conditional mutual information between the fea-tures. Instead, a better predictor of naive Bayes ac-curacy is the amount of information about the class that is lost because of the independence assump-tion.
Article
Full-text available
This paper develops an order-up-to S inventory model that is designed to handle multiple items, resource constraints, lags in delivery, and lost sales without sacrificing computational simplicity. Mild conditions are shown to ensure that the expected average holding cost and the expected average shortage cost are separable convex functions of the order-up-to levels. We develop nonparametric estimates of these costs and use them in conjunction with linear programming to produce what is termed the "LP policy." The LP policy has two major advantages over traditional methods: first, it can be computed in complex environments such as the one described above; and second, it does not require an explicit functional form of demand, something that is difficult to specify accurately in practice. In two numerical experiments designed so that optimal policies could be computed, the LP policy fared well, differing from the optimal profit by an average of 2.20% and 1.84%, respectively. These results compare quite favorably with the errors incurred in traditional methods when a correctly specified distribution uses estimated parameters. Our findings support the effectiveness of this mathematical programming technique for approximating complex, real-world inventory control problems.
Article
Full-text available
This paper considers a retailer inventory system with N first-echelon stores and a single second-echelon distribution center (DC). Customer demands at the stores are assumed to be random. When a stockout occurs, customers are willing to wait for their order to be filled with a known probability. Customers who are unwilling to wait result in lost sales. The first and second echelons are both restocked at fixed, equally spaced time points, where the store restocking frequency is an integer multiple of the DC restocking frequency. We also assume that replenishment quantities at both echelons can be adjusted up to the time of delivery, resulting in replenishment lead times equal to zero. This simplification allows us to determine optimal solutions for the partial lost sales case, which has proven intractable for two-echelon formulations with lead times. Computational results are given for illustrative examples.
Article
Full-text available
The occurrence of temporary stock-outs at retail is common in frequently purchased product categories. Available empirical evidence suggests that when faced with stock-outs, consumers are often willing to buy substitute items. An important implication of this consumer behavior is that observed sales of an item no longer provide a good measure of its core demand rate. Sales of items that stock-out are right-censored, while sales of other items are inflated because of substitutions. Knowledge of the true demand rates and substitution rates is important for the retailer for a variety of category management decisions such as the ideal assortment to carry, how much to stock of each item, and how often to replenish the stock. The estimated substitution rates can also be used to infer patterns of competition between items in the category. In this paper we propose methods to estimate demand rates and substitution rates in such contexts. We develop a model of customer arrivals and choice between goods that explicitly allows for possible product substitution and lost sales when a customer faces a stock-out. The model is developed in the context of retail vending, an industry that accounts for a sizable part of the retail sales of many consumer products. We consider the information set available from two kinds of inventory tracking systems. In the best case scenario of a perpetual inventory system in which times of stock-out occurrence and cumulative sales of all goods up to these times are observed, we derive Maximum Likelihood Estimates (MLEs) of the demand parameters and show that they are especially simple and intuitive. However, state-of-the-art inventory systems in retail vending provide only periodic data, i.e., data in which times of stock-out occurrence are unobserved or “missing.” For these data we show how the Expectation-Maximization (EM) algorithm can be employed to obtain the MLEs of the demand parameters by treating the stock-out times as missing data. We show an application of the model to daily sales and stocking data pooled across multiple beverage vending machines in a midwestern U.S. city. The vending machines in the application carry identical assortments of six brands. Since the number of parameters to be estimated is too large given the available data, we discuss possible restrictions of the consumer choice model to accomplish the estimation. Our results indicate that demand rates estimated naively by using observed sales rates are biased, even for items that have very few occurrences of stock-outs. We also find significant differences among the substitution rates of the six brands. The methods proposed in our paper can be modified to apply to many nonvending retail settings in which consumer choices are observed, not their preferences, and choices are constrained because of unavailability of items in the choice set. One such context is in-store grocery retailing, where similar issues of information availability arise. In this context an important issue that would need to be dealt with is changes in the retail environment caused by retail promotions.
Conference Paper
Full-text available
Several algorithms have been proposed for finding the "best," "opti- mal," or "most interesting" rule(s) in a database according to a vari- ety of metrics including confidence, support, gain, chi-squared value, gini, entropy gain, laplace, lift, and conviction. In this paper, we show that the best rule according to any of these metrics must reside along a support/confidence border. Further, in the case of conjunctive rule mining within categorical data, the number of rules along this border is conveniently small, and can be mined efft- ciently from a variety of real-world data-sets. We also show how this concept can be generalized to mine all rules that are best according to any of these criteria with respect to an arbitrary subset of the population of interest. We argue that by returning a broader set of rules than previous algorithms, our techniques allow for improved insight into the data and support more user-interaction in the optimized rule-mining process.
Article
Full-text available
Storing and using specific instances improves the performance of several supervised learning algorithms. These include algorithms that learn decision trees, classification rules, and distributed networks. However, no investigation has analyzed algorithms that use only specific instances to solve incremental learning tasks. In this paper, we describe a framework and methodology, called instance-based learning, that generates classification predictions using only specific instances. Instance-based learning algorithms do not maintain a set of abstractions derived from specific instances. This approach extends the nearest neighbor algorithm, which has large storage requirements. We describe how storage requirements can be significantly reduced with, at most, minor sacrifices in learning rate and classification accuracy. While the storage-reducing algorithm performs well on several real-world databases, its performance degrades rapidly with the level of attribute noise in training instances. Therefore, we extended it with a significance test to distinguish noisy instances. This extended algorithm's performance degrades gracefully with increasing noise levels and compares favorably with a noise-tolerant decision tree algorithm.
Article
Full-text available
Recent work in supervised learning has shown that a surprisingly simple Bayesian classifier with strong assumptions of independence among features, called naive Bayes, is competitive with state-of-the-art classifiers such as C4.5. This fact raises the question of whether a classifier with less restrictive assumptions can perform even better. In this paper we evaluate approaches for inducing classifiers from data, based on the theory of learning Bayesian networks. These networks are factored representations of probability distributions that generalize the naive Bayesian classifier and explicitly represent statements about independence. Among these approaches we single out a method we call Tree Augmented Naive Bayes (TAN), which outperforms naive Bayes, yet at the same time maintains the computational simplicity (no search involved) and robustness that characterize naive Bayes. We experimentally tested these approaches, using problems from the University of California at Irvine repository, and compared them to C4.5, naive Bayes, and wrapper methods for feature selection.
Chapter
Full-text available
We are given a large database of customer transactions. Each transaction consists of items purchased by a customer in a visit. We present an efficient algorithm that generates all significant association rules between items in the database. The algorithm incorporates buffer management and novel estimation and pruning techniques. We also present results of applying this algorithm to sales data obtained from a large retailing company, which shows the effectiveness of the algorithm.
Article
Full-text available
Naive-Bayes induction algorithms were previously shown to be surprisingly accurate on many classification tasks even when the conditional independence assumption on which they are based is violated. However, most studies were done on small databases. We show that in some larger databases, the accuracy of Naive-Bayes does not scale up as well as decision trees. We then propose a new algorithm, NBTree, which induces a hybrid of decision-tree classifiers and Naive-Bayes classifiers: the decision-tree nodes contain univariate splits as regular decision-trees, but the leaves contain Naive-Bayesian classifiers. The approach retains the interpretability of Naive-Bayes and decision trees, while resulting in classifiers that frequently outperform both constituents, especially in the larger databases tested.
Article
Full-text available
A training algorithm that maximizes the margin between the training patterns and the decision boundary is presented. The technique is applicable to a wide variety of classifiaction functions, including Perceptrons, polynomials, and Radial Basis Functions. The effective number of parameters is adjusted automatically to match the complexity of the problem. The solution is expressed as a linear combination of supporting patterns. These are the subset of training patterns that are closest to the decision boundary. Bounds on the generalization performance based on the leave-one-out method and the VC-dimension are given. Experimental results on optical character recognition problems demonstrate the good generalization obtained when compared with other learning algorithms. 1
Article
Full-text available
. We evaluate the power of decision tables as a hypothesis space for supervised learning algorithms. Decision tables are one of the simplest hypothesis spaces possible, and usually they are easy to understand. Experimental results show that on artificial and real-world domains containing only discrete features, IDTM, an algorithm inducing decision tables, can sometimes outperform state-of-the-art algorithms such as C4.5. Surprisingly, performance is quite good on some datasets with continuous features, indicating that many datasets used in machine learning either do not require these features, or that these features have few values. We also describe an incremental method for performing crossvalidation that is applicable to incremental learning algorithms including IDTM. Using incremental cross-validation, it is possible to cross-validate a given dataset and IDTM in time that is linear in the number of instances, the number of features, and the number of label values. The time for incre...
Article
Many companies have automated their inventory management processes and now rely on information systems when making critical decisions. However, if the information is inaccurate, the ability of the system to provide a high availability of products at the minimal operating cost can be compromised. In this paper, analytical and simulation modelling demonstrate that even a small rate of stock loss undetected by the information system can lead to inventory inaccuracy that disrupts the replenishment process and creates severe out-of-stock situations. In fact, revenue losses due to out-of-stock situations can far outweigh the stock losses themselves. This sensitivity of the performance to the inventory inaccuracy becomes even greater in systems operating in lean environments. Motivated by an automatic product identification technology under development at the Auto-ID Center at MIT, various methods of compensating for the inventory inaccuracy are presented and evaluated. Comparisons of the methods reveal that the inventory inaccuracy problem can be effectively treated even without automatic product identification technologies in some situations.
Article
An important component of many data mining projects is finding a good classification algorithm, a process that requires very careful thought about experimental design. If not done very carefully, comparative studies of classification and other types of algorithms can easily result in statistically invalid conclusions. This is especially true when one is using data mining techniques to analyze very large databases, which inevitably contain some statistically unlikely data. This paper describes several phenomena that can, if ignored, invalidate an experimental comparison. These phenomena and the conclusions that follow apply not only to classification, but to computational experiments in almost any aspect of data mining. The paper also discusses why comparative analysis is more important in evaluating some types of algorithms than for others, and provides some suggestions about how to avoid the pitfalls suffered by many experimental studies.
Article
Random forests are a combination of tree predictors such that each tree depends on the values of a random vector sampled independently and with the same distribution for all trees in the forest. The generalization error for forests converges a.s. to a limit as the number of trees in the forest becomes large. The generalization error of a forest of tree classifiers depends on the strength of the individual trees in the forest and the correlation between them. Using a random selection of features to split each node yields error rates that compare favorably to Adaboost (Y. Freund & R. Schapire, Machine Learning: Proceedings of the Thirteenth International conference, ***, 148–156), but are more robust with respect to noise. Internal estimates monitor error, strength, and correlation and these are used to show the response to increasing the number of features used in the splitting. Internal estimates are also used to measure variable importance. These ideas are also applicable to regression.
Book
Artificial intelligence and expert systems have seen a great deal of research in recent years, much of which has been devoted to methods for incorporating uncertainty into models. This book is devoted to providing a thorough and up-to-date survey of this field for researchers and students.
Article
We propose a network architecture which uses a single internal layer of locally-tuned processing units to learn both classification tasks and real-valued function approximations (Moody and Darken 1988). We consider training such networks in a completely supervised manner, but abandon this approach in favor of a more computationally efficient hybrid learning method which combines self-organized and supervised learning. Our networks learn faster than backpropagation for two reasons: the local representations ensure that only a few units respond to any given input, thus reducing computational overhead, and the hybrid learning rules are linear rather than nonlinear, thus leading to faster convergence. Unlike many existing methods for data analysis, our network architecture and learning rules are truly adaptive and are thus appropriate for real-time use.
Article
2,810 consumers were interviewed regarding their response to 5 items (removed from the shelves by the researchers) that were out of stock. 375 Ss could not find the specific items they wanted. 32% of these Ss purchased a different brand, 41% purchased a different size or variety of the same brand, 13% delayed purchase, and 14% went to another store. Perceived product risk, urgency of need, and produce usage were factors in the decision to substitute. Repeat brand purchase patterns did not influence the decision to substitute but did influence the specific substitution made, if an item was substituted for what was out of stock. In those cases where no substitution was made, store loyalty influenced the decision to delay the purchase or to go to another store. (PsycINFO Database Record (c) 2012 APA, all rights reserved)
Chapter
This paper presents a generalization of the perception learning procedure for learning the correct sets of connections for arbitrary networks. The rule, falled the generalized delta rule, is a simple scheme for implementing a gradient descent method for finding weights that minimize the sum squared error of the sytem's performance. The major theoretical contribution of the work is the procedure called error propagation, whereby the gradient can be determined by individual units of the network based only on locally available information. The major empirical contribution of the work is to show that the problem of local minima not serious in this application of gradient descent. Keywords: Learning; networks; Perceptrons; Adaptive systems; Learning machines; and Back propagation
Article
Vendor-managed inventory (VMI) is a supply-chain initiative where the supplier is authorized to manage inventories of agreed-upon stock-keeping units at retail locations. The benefits of VMI are well recognized by successful retail businesses such as Wal-Mart. In VMI, distortion of demand information (known as bullwhip effect) transferred from the downstream supply-chain member (e.g., retailer) to the upstream member (e.g., supplier) is minimized, stockout situations are less frequent, and inventory-carrying costs are reduced. Furthermore, a VMI supplier has the liberty of controlling the downstream resupply decisions rather than filling orders as they are placed. Thus, the approach offers a framework for synchronizing inventory and transportation decisions. In this paper, we present an analytical model for coordinating inventory and transportation decisions in VMI systems. Although the coordination of inventory and transportation has been addressed in the literature, our particular problem has not been explored previously. Specifically, we consider a vendor realizing a sequence of random demands from a group of retailers located in a given geographical region. Ideally, these demands should be shipped immediately. However, the vendor has the autonomy of holding small orders until an agreeable dispatch time with the expectation that an economical consolidated dispatch quantity accumulates. As a result, the actual inventory requirements at the vendor are partly dictated by the parameters of the shipment-release policy in use. We compute the optimum replenishment quantity and dispatch frequency simultaneously. We develop a renewaltheoretic model for the case of Poisson demands, and present analytical results.
Article
After discovering the inadequacy of traditional relationships across the value‐chain, many organizations today are establishing new forms of interorganizational systems with their suppliers and customers in an effort to improve total channel performance. Electronic commerce linkages are being created between independent organizations in multiple industries, including manufacturing, financial services, transportation, and retailing. Efforts to improve channel efficiency using business‐to‐business electronic commerce systems generally require increased interdependence and expanded coordination between independent firms to capture the potential benefits enabled by tighter interorganizational integration of operations. Participating firms can gain dramatic benefits from establishing electronic linkage only when the system is used to increase interdependence and to expand coordination between firms involved in the new interorganizational relationship. Drawing on theoretical and empirical research on electronic communications and inter‐firm designs, we develop and test a model for the relationship between performance, interdependence and coordination of firms involved in interorganizational relations within the US grocery channel. The research design includes qualitative case study analysis and quantitative survey data analysis to validate the key case study findings. Both qualitative and quantitative findings indicate that channel performance, interdependence, and coordination are closely related for firms in interorganizational relationships.
Article
This paper investigates consumer reactions to out-of-stocks (OOS)—which are unexpected and temporary in nature—as opposed to permanent assortment reductions (PAR). We discuss similarities and differences between OOS and PAR reactions as well as their underlying factors, and empirically test our propositions in two product categories. Our results indicate that retailer losses incurred in case of a PAR may be substantially larger than those in case of a stock-out for the same item. This suggests that retailers (i) should be cautious in extrapolating consumer response to stock-outs to PAR situations, (ii) should keep a minimum of variation in the assortment even if this implies a higher stock-out risk, (iii) may wish to explicitly signal the temporary character of a stock-out by leaving the shelf space of the OOS item unused. The results further suggest that stock-out losses may disproportionately grow with OOS frequency and duration, emphasizing the need to keep their occurrence and length within limits.
Article
Shelf space is one of the most important resources of a retail firm. This paper formulates a model and proposes an approach which is similar to the algorithm used for solving a knapsack problem. Subject to given constraints, the proposed heuristic allocates shelf space item by item according to a descending order of sales profit for each item per display area or length. Through the use of simulations, the performances of objective value and the computational efficiency of this method are evaluated. Three options are also proposed for improving the heuristics. Compared to an optimal method, the improved heuristic is shown to be a very efficient algorithm which allocates shelf space at near-optimal levels.
Article
Classification systems play an important role in business decision-making tasks by classifying the available information based on some criteria. The objective of this research is to assess the relative performance of some well-known classification methods. We consider classification techniques that are based on statistical and AI techniques. We use synthetic data to perform a controlled experiment in which the data characteristics are systematically altered to introduce imperfections such as nonlinearity, multicollinearity, unequal covariance, etc. Our experiments suggest that data characteristics considerably impact the classification performance of the methods. The results of the study can aid in the design of classification systems in which several classification methods can be employed to increase the reliability and consistency of the classification.
Article
Manufacturers as well as retailers can suffer important losses as a result of stock-outs. The magnitude of these losses depends on specific consumer reactions, which have been found to vary with product, consumer, and situation factors. This paper presents a conceptual framework that integrates the major determinants of consumer reactions to stock-outs. The theoretical relationships provide explanations for the marked differences in stock-out effects observed in previous studies. Moreover, the framework can be empirically implemented, allowing retailers and manufacturers to determine how much each factor contributes to stock-out losses. We collect survey data to provide evidence on the relevance of the framework and the direction and importance of the effect of different consumer behaviors.
Article
In this paper, we provide an integrated framework for portfolio selection, which is adaptable to the needs of financial organizations and individual investors, and as an organized approach of selecting efficient portfolios for investments. We focus our discussion on the implementation of this framework for a Web-based Decision Support System (DSS) based on our prototype named WPSS—A Web-based Portfolio Selection System for Chinese financial markets. In this system, we adopt technologies such as online analytical processing as an add-on tool for analytical purpose, as well as using Parallel Virtual Machine (PVM) to improve overall performance.
Article
One of the most challenging problems in data mining is to develop scalable algorithms capable of mining massive data sets whose sizes exceed the capacity of a computer's memory. In this paper, we propose a new decision tree algorithm, named SURPASS (for Scaling Up Recursive Partitioning with Sufficient Statistics), that is highly effective in handling such large data. SURPASS incorporates linear discriminants into decision trees' recursive partitioning process. In SURPASS, the information required to build a decision tree is summarized into a set of sufficient statistics, which can be gathered incrementally from the data, by reading a subset of the data from storage space to main memory one at a time. As a result, the data size that can be handled by this algorithm is independent of memory size. We apply SURPASS to three large data sets pertaining to pattern recognition and intrusion detection problems. The results indicate that SURPASS scales up well against large data sets and produces decision tree models with very high quality.
Article
A retail chain manager must draw on experience based on data available from his points of sale to diagnose space misallocations in stores and to make recommendations. This paper presents an empirical estimate of shelf space elasticities from a variety store chain database at product category level with a share of space vs. share of sales econometric model. It suggests that external influences could explain space elasticity differences. Results show that space elasticities increase with the impulse buying rate of the product category and do not depend on the type of store.
Article
In these days, EC companies are eager to learn about their customers using data mining technologies. But the diverse situations of such companies make it difficult to know which is the most effective algorithm for the given problems. Recently, a movement towards combining multiple classifiers has emerged to improve classification results. In this paper, we propose a method for the prediction of the EC customer's purchase behavior by combining multiple classifiers based on genetic algorithm. The method was tested and evaluated using Web data from a leading EC company. We also tested the validity of our approach in general classification problems using handwritten numerals. In both cases, our method shows better performance than individual classifiers and other known combining methods we tried.
Conference Paper
This paper presents PAC-learning analyses for instance-based learning algorithms for both symbolic and numeric-prediction tasks. The algorithms analyzed employ a variant of the k-nearest neighbor pattern classifier. The main results of these analyses are that the IB1 instance-based learning algorithm can learn, using a polynomial number of instances, a wide range of symbolic concepts and numeric functions. In. addition, we show that a bound on the degree of difficulty of predicting symbolic values may be obtained by considering the size of the boundary of the target concept, and a bound on the degree of difficulty in predicting numeric values may be obtained by considering the maximum absolute value of the slope between instances in the instance space. Moreover, the number of training instances required by IB1 is polynomial in these parameters. The implications of these results for the practical application of instance-based learning algorithms are discussed.
Article
Competition and globalization imply a very accurate production and sourcing management of the Textile–Apparel–Distribution network actors. A sales forecasting system is required to respond to the versatile textile market and the needs of the distributors. Nowadays, due to the specific constraints of the textile sales (numerous and new items, short life time), existing forecasting models are generally unsuitable or unusable. We propose a forecasting system, based on clustering and classification tools, which performs mid-term forecasting. Performances of our models are evaluated using real data from an important French textile distributor.
Article
Consider a model whose structure is representable via recursive graph or Bayes network and in which both observable and unobservable variables are involved. Assume that the variables are all binary and suppose that we want to predict about the unobservables based on the evidence from the observables. In the real world, we may not be able construct a model that exactly explains a certain phenomenon we are interested in. However, if we may give predictions in terms of class or category rather than probability, then we may not have to know the exact details of the model. It is shown in this article that when the observables are conditionally independent and conditionally stochastically ordered given the unobservables, the unobservables are conditionally ordered given the observables. This result suggests to some extent robustness of the conditional probabilities of the observables given the unobservables, and the simulation result strongly supports the suggestion.
Article
A study of more than 600 retail outlets in Health Care finds that stock-outs are far more costly than most companies imagine.
Article
Tree induction methods and linear models are popular techniques for supervised learning tasks, both for the prediction of nominal classes and numeric values. For predicting numeric quantities, there has been work on combining these two schemes into `model trees', i.e. trees that contain linear regression functions at the leaves. In this paper, we present an algorithm that adapts this idea for classification problems, using logistic regression instead of linear regression. We use a stagewise fitting process to construct the logistic regression models that can select relevant attributes in the data in a natural way, and show how this approach can be used to build the logistic regression models at the leaves by incrementally refining those constructed at higher levels in the tree. We compare the performance of our algorithm to several other state-of-the-art learning schemes on 36 benchmark UCI datasets, and show that it produces accurate and compact classifiers. This is an author’s version of an article published on the journal: Machine Learning. The original publication is available at www.springerlink.com.
Article
SUMMARY The problem of discrimination when all or most of the observations are qualitative is discussed. The method of logistic discrimination introduced by Cox (1966) and Day & Kerridge (1967) is extended to the situation where separate samples are taken from each population, using the results of Aitchison & Silvey (1958) on constrained maximum likelihood estimation. The method is further extended to discrimination between three or more populations. The properties of logistic discrimination are investigated by simulation and the method is applied to the differential diagnosis of kerato-conjundivitis sicca.
Article
DOI:10.1016/j.dss.2005.05.006 In this paper, we present the findings of a case study on the development of a radio frequency identification (RFID) prototype system that is integrated with mobile commerce (m-commerce) in a container depot. A system architecture capable of integrating mobile commerce and RFID applications is proposed. The system architecture is examined and explained in the context of the case study. The aims of the system are to (i) keep track of the locations of stackers and containers, (ii) provide greater visibility of the operations data, and (iii) improve the control processes. The case study illustrates the benefits and advantages of using an RFID system, particularly its support of m-commerce activities in the container depot, and describes some of the most important problems and issues. Finally, several research issues and directions of RFID applications in container depots are presented and discussed.
Conference Paper
Predicting rare classes effectively is an important problem. The definition of effective classifier, embodied in the classifier evaluation metric, is however very subjective, dependent on the application domain. In this paper a wide variety of point-metrics are put into a common analytical context defined by the recall and precision of the target rare class. This enables us to compare various metrics in an objective, domain-independent manner. We judge their suitability for the rare class problems along the dimensions of learning difficulty and levels of rarity. This yields many valuable insights. In order to address the goal of achieving better recall and precision, we also propose a way of comparing classifiers directly based on the relationships between recall and precision values. It resorts to a composite point-metric only when recall-precision based comparisons yield conflicting results.
Article
. An important component of many data mining projects is finding a good classification algorithm, a process that requires very careful thought about experimental design. If not done very carefully, comparative studies of classification and other types of algorithms can easily result in statistically invalid conclusions. This is especially true when one is using data mining techniques to analyze very large databases, which inevitably contain some statistically unlikely data. This paper describes several phenomena that can, if ignored, invalidate an experimental comparison. These phenomena and the conclusions that follow apply not only to classification, but to computational experiments in almost any aspect of data mining. The paper also discusses why comparative analysis is more important in evaluating some types of algorithms than for others, and provides some suggestions about how to avoid the pitfalls suffered by many experimental studies. Keywords: classification, comparative studi...
Article
Modern retailers typically manage thousands of individual stock keeping units (skus) and are faced with the complex task of stocking, pricing, promoting and maintaining an appropriate product assortment. This task is further complicated by the substantial heterogeneity in consumer preference for different product offerings. In this environment, it is critical that retail management has some understanding of how consumers respond to product unavailability. The goal of this paper is to develop an understanding of the process of consumer response to product unavailability (stockouts). We attack this problem using two disparate methods. First, we conduct a controlled laboratory experiment to gain insight into what consumers perceive to be the "cost" of choosing from a restricted set of product alternatives. We nd that not only do consumers react negatively to the omission of preferred brands, but also that they experience dissatisfaction when previously available (but unchosen) bra...
Article
The use of entropy as a distance measure has several benefits. Amongst other things it provides a consistent approach to handling of symbolic attributes, real valued attributes and missing values. The approach of taking all possible transformation paths is discussed. We describe K*, an instance-based learner which uses such a measure, and results are presented which compare favourably with several machine learning algorithms. Introduction The task of classifying objects is one to which researchers in artificial intelligence have devoted much time and effort. The classification problem is hard because often the data available may be noisy or have irrelevant attributes, there may be few examples to learn from or simply because the domain is inherently difficult. Many different approaches have been tried with varying success. Some well known schemes and their representations include: ID3 which uses decision trees (Quinlan 1986), FOIL which uses rules (Quinlan 1990), PROTOS which is a case...
Article
The application of boosting procedures to decision tree algorithms has been shown to produce very accurate classifiers. These classifiers are in the form of a majority vote over a number of decision trees. Unfortunately, these classifiers are often large, complex and difficult to interpret. This paper describes a new type of classification rule, the alternating decision tree, which is a generalization of decision trees, voted decision trees and voted decision stumps. At the same time classifiers of this type are relatively easy to interpret.
Approaches to Measuring On-Shelf Availability at the Point of
  • G Hausruckinger
G. Hausruckinger, Approaches to Measuring On-Shelf Availability at the Point of Sale, ECR Europe, White Paper,, 2006.