Avoiding food waste from restaurant tickets: a big data management tool

Abstract

Purpose
This study aims to address the global food waste problem in restaurants by analyzing customer sales information provided by restaurant tickets to gain valuable insights into directing sales of perishable products and optimizing product purchases according to customer demand.

Design/methodology/approach
A system based on unsupervised machine learning (ML) data models was created to provide a simple and interpretable management tool. This system performs analysis based on two elements: first, it consolidates and visualizes mutual and nontrivial relationships between information features extracted from tickets using multicomponent analysis, bootstrap resampling and ML domain description. Second, it presents statistically relevant relationships in color-coded tables that provide food waste-related recommendations to restaurant managers.

Findings
The study identified relationships between products and customer sales in specific months. Other ticket elements have been related, such as products with days, hours or functional areas and products with products (cross-selling). Big data (BD) technology helped analyze restaurant tickets and obtain information on product sales behavior.

Research limitations/implications
This study addresses food waste in restaurants using BD and unsupervised ML models. Despite limitations in ticket information and lack of product detail, it opens up research opportunities in relationship analysis, cross-selling, productivity and deep learning applications.

Originality/value
The value and originality of this work lie in the application of BD and unsupervised ML technologies to analyze restaurant tickets and obtain information on product sales behavior. Better sales projection can adjust product purchases to customer demand, reducing food waste and optimizing profits.

Full text

Avoiding food waste from
restaurant tickets: a big data
management tool
Ismael G
omez-Talal
Department of Signal Theory and Communications and Telematic Systems and
Computation, Rey Juan Carlos University –Fuenlabrada Campus,
Fuenlabrada, Spain
Lydia Gonz
alez-Serrano
Department of Business and Management,
Rey Juan Carlos University –Fuenlabrada Campus, Fuenlabrada, Spain
Jos
e Luis Rojo-Álvarez
Department of Signal Theory and Communications and Telematic Systems and
Computation, Rey Juan Carlos University –Fuenlabrada Campus,
Fuenlabrada, Spain, and
Pilar Tal
on-Ballestero
Department of Business and Management,
Rey Juan Carlos University –Fuenlabrada Campus, Fuenlabrada, Spain
Abstract
Purpose –This study aims to address the global food waste problem in restaurants by analyzing customer
sales information provided by restaurant tickets to gain valuable insights into directing sales of perishable
products and optimizing product purchasesaccording to customer demand.
Design/methodology/approach –A system based on unsupervised machine learning (ML) data models
was created to provide a simple and interpretable management tool. This system performs analysis based on
two elements: first, it consolidates and visualizes mutual and nontrivial relationships between information
features extracted from tickets using multicomponent analysis, bootstrap resampling and ML domain
description. Second, it presents statistically relevant relationships in color-coded tables that provide food
waste-related recommendations to restaurant managers.
Findings –The study identified relationships between products and customer sales in specific months.
Other ticket elements have been related, such as products with days, hours or functional areas and products
with products (cross-selling). Big data (BD) technology helped analyze restaurant tickets and obtain
information on product sales behavior.
Research limitations/implications –This study addresses food waste in restaurants using BD
and unsupervised ML models. Despite limitations in ticket information and lack of product detail, it
opens up research opportunities in relationship analysis, cross-selling, productivity and deep learning
applications.
The authors would like to especially thank Dynameat for providing the data used in this work and
for the useful discussions.
Funding: This work was partly supported by the State Research Agency of the Ministry of Science
and Innovation with reference code AEI/10.13039/501100011033 and PID2022-140786NB-C31.
Statements and declarations. The authors declare no conflict of interest.
Avoiding food
waste
Received 15 January2023
Revised 3 July 2023
26 August 2023
Accepted 30 August2023
Journal of Hospitality and
Tourism Technology
© Emerald Publishing Limited
1757-9880
DOI 10.1108/JHTT-01-2023-0012
The current issue and full text archive of this journal is available on Emerald Insight at:
https://www.emerald.com/insight/1757-9880.htm

Originality/value –The value and originality of this work lie in the application of BD and unsupervised
ML technologies to analyze restaurant tickets and obtain information on product sales behavior. Better sales
projection can adjustproduct purchasesto customer demand, reducing food waste and optimizing profits.
Keywords Big data, Food waste, Hospitality industry, Unsupervised learning, Bootstrap resampling,
Sales forecasting
Paper type Research paper
从餐厅票据中避免食物浪费：一个大数据管理工具
摘要
目的 –本研究旨在通过分析餐厅票据上的顾客销售信息,解决全球餐厅食物浪费问题,获取有价值的
洞察,指导易腐产品的销售,并根据顾客需求优化产品采购。
设计/方法/途径–创建了一个基于无监督机器学习（ML）数据模型的系统,提供一个简单且易于解
释的管理工具。该系统首先整合并可视化票据信息特征之间的相互关系和非平凡关系,使用多组分分
析、自助重采样和ML领域描述。其次,它以彩色编码表格形式展现统计学上显著的关系,为餐厅管理
者提供与食物浪费相关的建议。
发现–研究识别了特定月份内产品与顾客销售之间的关系。还相关联了其他票据元素,例如产品与
天、小时或功能区域的关系,以及产品与产品之间的交叉销售（cross-selling）。大数据（BD）技术
有助于分析餐厅票据,获取产品销售行为的信息。
研究局限性/影响–尽管票据信息存在局限性,且缺乏产品细节,但本研究为关系分析、交叉销售、
生产力和深度学习应用等领域开辟了研究机会。
原创性/价值–本工作的原创性在于应用BD和无监督ML技术分析餐厅票据,获取产品销售行为的信
息。更好的销售预测可以调整产品采购以满足顾客需求,减少食物浪费,优化利润。
关键词大数据食品浪费酒店业无监督学习自助抽样销售预测
文章类型研究型论文
1. Introduction
The United Nations’2030 Agenda highlights sustainable development goals related to
sustainable sales patterns and production chains. Countries like France, Italy and Spain
have implemented legislation to curb food waste. In 2006, EU-27 countries wasted 12 million
tons of restaurant food (Ghosh et al.,2016), while the US restaurant industry wasted food
valued at $25bn annually (Srijuntrapun et al., 2022). Studies suggest 75% of this waste could
be prevented with efficient management (Engström and Carlsson-Kanyama, 2004). Current
solutions emphasize efficient restaurant management and forecasting to reduce waste
(Mond
ejar-Jim
enez et al., 2016;Young et al.,2018). Forecasting is crucial for revenue
management (RM), which successfully manages perishable goods under demand
uncertainty and maximizes profit by minimizing waste (Gonz
alez-Serrano and Tal
on-
Ballestero, 2020;Kimes et al.,1998). RM system helps to reduce wasted inventory and, thus,
increases profit. The adoption of RM tends to improve yields because scarce resources are
more effectively managed, and the waste of missed revenue is avoided (Weatherford and
Kimes, 2003).
Restaurants usually need more detailed customer data than hotels, limiting their ability
to analyze consumer behavior (Cavusoglu, 2019). They rely on sales tickets, which lack
specific customer details but offer insights into buying patterns. Traditional forecasting is
often insufficient for tackling food waste (Rizou et al.,2020). However, big data (BD)
technologies allow for better analysis of sales tickets, offering a deep dive into customer
behavior and improved management practices that curb waste (Meek et al.,2021;Samara
et al.,2020;Tao et al.,2020). BD is also instrumental in sales forecasting (Schmidt et al., 2022;
Nadkarni et al.,2019;Sakib, 2023).
JHTT

Forecasting sales are pivotal for the hospitality industry (Mariani et al., 2018;Hsu et al., 2022).
While many models exist, a significant challenge remains, namely, restaurants usually need
more specific customer details and rely on point of sale (POS) systems. Existing research has
used linear and nonlinear models for forecasting (Bang et al., 2019;Tanizaki et al., 2019;
Priyadarshi et al., 2019), but there is a gap in the literature regarding machine learning (ML)-
based sales forecasting (Doborjeh et al., 2022). This paper introduces an innovative ML model
using multiple correspondence analysis (MCA) for sales forecasting. Accurate sales predictions
are essential for understanding customer behavior, ensuring efficient procurement, minimizing
waste and assisting decision-making (Tsoumakas, 2019). The research harnesses BD techniques
analyzing 367,527 restaurant tickets provided by Dynameat, aiming to increase profitability
using artificial intelligence (AI) models rooted in RM and menu engineering strategies.
The scheme of the paper is as follows. Section 2 gathers the literature regarding the
challenges and strategies of food waste in food service with particular attention to
forecasting and current methods using BD technology. Section 3 facilitates the equations of
the preprocessing as mentioned above blocks, such as MCA, bootstrap resampling on the
categories and the support vector domain descriptor for the computation of confidence
volumes. Section 4 exposes the experiments and results using the database of restaurant
tickets to study sales. Subsequently, Section 5 presents the main conclusions. Section 6
offers the theoretical implications and Section 7 practical implications of the unsupervised
restoration model. Finally, Section 8 shows the limitations and future research.
2. Literature review
2.1 Food waste reduction strategies in restaurants
The food service sector significantly contributes to food waste, accounting for a quarter of
consumer waste, with much of this waste being avoidable (Hennchen, 2019;Dhir et al., 2020).
Gunders and Bloom (2017) estimated that 4% to 10% of food is wasted due to consumer
behavior, making consumer sales analysis vital in reducing food waste in restaurants (Martin-
Rios et al., 2018). Food waste drivers can be categorized into internal, microenvironmental and
macroenvironmental drivers (Özbük and Coskun, 2020). Internal drivers include organization,
operational activities and marketing strategies, while microenvironmental drivers involve
competition, consumers and suppliers (Mirosa et al., 2018). Macroenvironmental factors relate
to political, ecosocial and natural influences (do Carmo Stangherlin and De Barcellos, 2018).
Operational activities like planning, purchasing, production and inventory management
are the primary causes of food waste (Özbük and Coskun, 2020). Inadequate planning,
stemming from inaccurate forecasting, poor demand planning or lack of menu planning, can
lead to over-purchasing and waste (Derqui et al.,2018;Pinto et al.,2018;McCray et al.,2018).
Accurate sales prediction of menu items is essential for environmental sustainability by
reducing preconsumer food waste and for the profitability of restaurants by ensuring proper
food stock ordering (Posch et al., 2022). Most restaurants use electronic POS systems, which
collect valuable data for forecasting. Despite challenges in using this data due to complex
supply chains in retail, the restaurant industry’s short lead times and more straightforward
supply chain make it viable to analyze customer sales information through POS systems
and BD technology. This enables predictions of perishable product sales and optimization of
purchases according to customer demand.
2.2 Sales forecasting in restaurants
Restaurant sales forecasting has historically depended on intuitive techniques based on
managerial experience. However, this approach must often be revised given the myriad factors
affecting sales, such as time, weather and economic conditions (van der Vorst et al.,1998). More
Avoiding food
waste

accurate forecasting is pivotal for effective production and inventory management (Goonan
et al., 2014). Although many in the industry still rely on these judgmental techniques (Lasek
et al.,2016), ML offers an unbiased, adaptive solution for deriving forecasts from sales data
(Tsoumakas, 2019).
Linear models, such as autoregressive integrated moving average and its variants, have
traditionally been used for time-series data in economics and tourism (Song et al., 2019;Bang
et al., 2019). In recent times, supervised learning has been gaining traction in sales
forecasting. These models use training and test data to predict specific variables and can be
classified into regression or classification models (Cunningham et al., 2008). However,
supervised models have challenges, such as complex hyperparameter adjustments and
slower training with large data sets (Vargas-Calder
on et al., 2021;Janikow, 1993).
In contrast, unsupervised learning focuses on discovering data patterns without prelabeled
outcomes, offering insights into potential data classifications. This approach has been used in
the hospitality sector’s customer relationship management, using methods like MCA to analyze
guest behavior (Dursun and Caber, 2016;van Leeuwen and Koole, 2022). BD and its integration
with statistical tools hold transformative potential, especially in hospitality and they aid in
understanding consumer needs, predicting trends and innovating business models (Shilo et al.,
2020;Bote-Curiel et al., 2019;Kim and Lee, 2019;Ogbeide et al., 2020;Mariani, 2019;Hsu, 2021).
The existing research emphasizes extracting insights from restaurant ticket data using
unsupervised learning but with a statistical approach focused on reducing dimensionality.
3. Data analytics methods
3.1 Big data set and methodology overview
The database used was configured from the ticket information of a restaurant in Madrid. The
company that provided the data (Dynameat) supervised the access to the information by
extracting and filtering the most important and valuable variables from POS systems per table.
The database consisted of 367,527 tickets with 32 ticket fields recorded during 2019. An
example of the ticket information, where the 32 data variables are collected, that has been
extracted from the restaurant POS systems can be visualized in Figure 1. In the preprocessing
process, the family of “Service”categories was eliminated, which corresponded to the start and
end tags of the service and not to a product offered in the restaurant. In addition, perishable
products were differentiated because they were the most likely to cause an increase in food
waste. Specifically, from all available data, we focus only on the products (food and beverages
consumed in the restaurant as reflected in the tickets) and the months of sales. With these two
variables, we leveraged an information column labeled “product family.”This distinguishes
perishable products more straightforwardly from those that are not. The filtering of perishable
products is based on the different selected families (“cheese”,“fish”,“meats”,“seafood”,
“vegetables and mushrooms”,“smoked and sauced”or “sausages”) following the consideration
of perishable products found in the literature (Terpstra et al., 2005). The objective is to filter the
product variables and the months of consumption to visualize the statistical relationships in a
three-dimensional latent space.
This work scrutinizes the statistical variability and the information distribution in
restaurant tickets using the MCA statistical method. On the one hand, the confidence
intervals of the weights of each MCA eigenvector are obtained, thus yielding a set of
informative directions sorted in terms of their relevance. On the other hand, based on this
information, confidence volumes are calculated for the categories projected to an observable
three-dimensional map given by the three most relevant directions, and these projections
allow us to visualize and quantify the statistical significance of the information relationships
across the data categories.
JHTT

The complexity of dealing with restoration data and reaching accurate conclusions leads us
to apply various techniques in this work. Figure 2 summarizes the research framework to
complete our proposed recommendation system.
3.2 Multiple correspondence analysis for feature description
The MCA techniques are used to compress categorical variables. Said techniques modify the
information of the variables by determining different weights in a projection matrix,
generating a compact and probabilistically founded visualization of the input data (Husson
and Josse, 2014). This methodallows a graphical representation of the relationship structure
of two or more categorical variables using positioning maps with probabilistic meaning.
Other authors define these techniques as implementing thewell-known principal component
analysis with categorical variables (Husson and Josse, 2014;Murakami, 2020). In this paper,
we present an implementation of the MCA algorithm that has been adapted to provide
statistical confidence descriptions in our data model by using bootstrap resampling to
scrutinize the variability of the observed categories when projected in the three-dimensional
space given by the three leading eigenvectors (Glynn, 2014).
In terms of mathematical notation, each n
th
categorical variable consists of Jkncategories,
so that X
N
n¼1
Jkn¼Jis the total number of categories in a data set. The binary matrix of
categories is defined in data matrix X, and by applying the correspondence analysis model,
the matrix of category weights is obtained. The sum of the total matrix Xby rows and
Figure 1.
An illustrative
example of ticket
information with
possible data stored
by the POS system
Avoiding food
waste

columns generates the Burt matrix (denoted as matrix M). Obtaining the probability matrix
Zis achieved by Z¼M1XX>, where vector ris defined as the number of rows of Zand
the parameter cas the number of columns of that same matrix. Eigenvalues and
eigenvectors are obtained by diagonalization of the centered probability matrix, as
explained in Gonz
alez-Serrano et al. (2020).
Figure 3(a) shows the eigenvalues of an example experiment using three categories,
where it can be reasoned that each k
th
component corresponds to the k
th
eigenvalue. The
three-dimensional probabilistic representation of the categories in Figure 3(b) allows us to
obtain factors and their relationships, considering the distances among each activation of
the empirical observations.
3.3 Bootstrap resampling for multiple correspondence analysis
Bootstrap resampling is a statistical technique used to estimate density distributions in
various statistical methods (Soguero-Ruiz et al., 2020). It involves creating a resampled
population by randomly selecting data points with replacements from the original sample.
The statistical measure of interest is then estimated on each resample, providing multiple
replications of the measure. This process allows us to estimate the density distribution of the
statistical measure that we are studying.
By applying the bootstrap resampling method in our study, we can determine the
inherent variability of different statistical descriptors in our data set. Specifically, we assess
the variability and statistical distribution of two measures in MCA. First, we estimate
confidence intervals for the weights of each eigenvector in the MCA projection. These
weights indicate the level of model complexity required to represent our data in low-
Figure 2.
Research framework
diagram of
preprocessing,
statistical model and
the application of the
recommendation
system
Load
dataset
Data
preprocessing
- Filtering perishables
and filter date (monthly)
- Filter categoricals
(product and monthly)
and one hot encoding
Build
MCA
Table of distances
Simplified table
Bootstrap
resampling
Eigenvectors
Eigenvectors
SVDD Method
Cloud SVDD
System
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
Source: Author’s own creation
JHTT

Figure 3.
Representation of the
eigenvalues of the
empirical samples (a)
and of the first three
empirical projected
categories in the
three-dimensional
latent space (b) on a
toy-data example
Avoiding food
waste

dimensional latent spaces (Hall and Martin, 1988). Second, we calculate confidence volumes
for categories when projected in this latent space, enabling us to compare the significance of
their co-occurrences (Carpenter and Bithell, 2000;Wood, 2005).
Regarding the mathematical notation used, the eigenvector matrix used in the MCA
model is here denoted as V, and the operator Hprovides the empirical eigenvector matrix
from the empirical data matrix. The input variable is then as follows:
V¼HX
ðÞ (1)
A resampled matrix X* is generated as a function of the size of the variable X(based on the
number of observations), with the number of rows being the size of the bootstrap resampling
value. The number of columns is the number of observations (Corral-De-Witt et al.,2019).
The representation of the projected categories in three dimensions is shown in Figure 4(a),
where we used the bootstrap resampling method to scrutinize the point estimate of the
dispersion of the categories.
The process is repeated Btimes (defined by the user), thus obtaining each resampled
matrix as X*(b), with b¼1, ...,B, and the resampled eigenvectors (Gonz
alez-Serrano et al.,
2020) are given by the following:
V*b
ðÞ
¼HX*b
ðÞ
 (2)
Now, we can readily represent the bootstrap histogram of each weight and each projected
category for each resample b.
3.4 Support vector domain description
While it is easy to provide a numerical description of confidence intervals in one-dimensional
statistical variables by determining lower and upper limits, establishing numerical limits for a
confidence volume is more complex. We used a widely used ML method called support vector
domain description (SVDD) for domain description to address this. By incorporating bootstrap
resampling, we determined that the overlap of confidence volumes is a criterion for determining
significant co-occurrence between two categories.
The SVDD method is a nonparametric procedure for estimating the domain support of an
arbitrary multidimensional statistical distribution (Tax and Duin, 2004). In our restaurant
ticket problem, the objective is to discern the bootstrap projected samples belonging to the
region of the confidence volume. It is possible to modify the width size parameter to adjust this
volume, thus allowing us to choose the desired number of support vectors by encompassing the
bootstrap-resampled activations in a hypersphere. Accordingly, the objective of the SVDD
method is to construct a hypersphere in high-dimensional feature space with the lowest
possible volume using a data set as ti
fg
N
i¼1of high dimension denoted as H,wheretheradiusR
is defined greater than zero and centered on a point defined as a(fulfilling the restriction a[H),
which in our case are the bootstrap resamples. The generated volume houses most of the
resampled projected categories (Tax and Duin, 2004).
Figure 4(b) shows that, in the above example of the resampling of the activations of the
observations of the samples of the three categories, it is possible to observe a confidence
volume using the SVDD method, where the bootstrap resamples that are considered within the
hypersphere are accumulated. Another tool used is the width of the Gaussian kernel from
the empirical probability density. This method is often used because it is based on the central
limit theorem (Liu et al., 2009). The interpretation of the categories’latent space via MCA is seen
as a probability mass function. In this context, centered samples represent larger values,
JHTT

indicating higher repetition within the statistical range. Conversely, less repetitive samples are
not centrally located. Regarding relationships between categories, samples closer to each other
indicate higher affinity. On the other hand, samples farther from each other imply lower
affinity. The conclusion of this example shows that categories 1 and 3 are statistically
Figure 4.
Representation of the
three main
eigenvectors by
bootstrap resampling
(a) and the projection
of the categories by
bootstrap resampling
together with their
confidence volumes
using the SVDD
technique (b)
Avoiding food
waste

significantly related, whereas category 2 is not because it is at a greater distance from the first
two categories.
4. Experiments and results
This section presents two experiments grounded on the unsupervised model and a final
experiment based on describing a product recommendation system. In the experiments, BD
filtering was performed to extract the perishable products fundamental to food waste
reduction.
4.1 Experiment on nonseasonal products
In the first experiment, the variables of perishable products and months are studied using
restaurant ticket data. Figure 1 represents a possible example of a restaurant ticket, showing
products and months, among other possible ticket fields. In this example, “Bluefin Tuna
Tartar”corresponds to one of the four product categories to be used in the feature vector,
and “March”is the month category for this ticket. In this case, additional filtering has been
carried out because the scrutinized products are consumed throughout the year. It is
necessary to filter the categories (following the statistical criterion of the sampling size) to
obtain a good-quality MCA model.
Thus, the sample size of perishable products is modified from 98 to 49 categories
(considered the minimum statistical sample size with a 95% confidence interval value).
Subsequently, the MCA model is built to calculate the distances between products and
months, thus generating a 49 12 matrix (number of perishable products by the number of
months). Figure 5 shows the distances of the empirical samples in the MCA feature space,
comparing the distances between the categories (perishable products and months). The table
uses an inverted heat map color scheme to highlight the relationships between different
categories. In this scheme, warm colors, such as deep reds and oranges, represent lower
values, indicating a stronger relationship between those categories. On the other hand,
higher values are presented with cool colors, such as soft shades of blue and green,
suggesting a lower relationship between these categories. The lower values indicate smaller
distances and a longer relationship between them. The coding color of the distance matrix is
based on the criteria of color map representations, where cool colors denote the longest
distances (and therefore the smallest ratio) and warm colors denote the values of greatest
interest (strongest ratios). Figure 5 shows the most relevant relationships obtained. For
example, the product of “1/2 tomato ventresca salad”has warm colors in July and September
with normalized distance values of 0.06 and 0.08, respectively (indicating a better
relationship between those two months). In the case of January, it has a cold color with a
normalized distance value of 0.75 (visualizing a low relationship in that month).
Regarding the analysis of the unsupervised MCA model, after running the experiment
with the study categories, the three eigenvectors shown in Figure 6(a) are obtained. The first
two eigenvectors exhibit many significant categories (indicated in red vertical stripes). In
contrast, the third eigenvector indicates that this model cannot detect significant categories.
This figure allows us to see how many features are significant in each eigenvector,
according to their confidence interval not overlapping zero. The red vertical lines emphasize
their visualization. The representation of the three primary eigenvectors, together with their
bootstrap confidence volumes, are shown in Figure 6(b), wherein the two main dimensions
of the cluster of categories appear. Therefore, the third eigenvector does not have significant
categories (red stripes), which distort the confidence volumes to a lesser extent. The number
of red vertical lines (indicating more significant categories) represents the clouds and,
therefore, a better visualization of the relationships.
JHTT

Figure 5.
Table of the empirical
sample distances of
the first experiment
Avoiding food
waste

To summarize the relationships between perishable products and months, a table has been
defined (products by months) like the table of distances, but in this case, strong relationships
are presented when they share the volume of confidence. In categories with a particular
spatial relationship using the MCA method, the histogram of the distances of the categories
has been studied. The median criterion defines the cut-off concerning the distances of the
relationships (between products and months) in green in the table. In this experiment,
Figure 7(a) shows the histogram of the distances and the median value (vertical line in red).
Based on this cut-off value, a simplified table has been designed, as shown in Figure 7(b),
where the same color criteria are followed (based on the heat maps). The red indicates a
strong relationship (where two clouds share the same space), and the green indicates
medium relationships (where they exceed the cut-off value but do not share the same cloud
space). Figure 7(a) shows the histogram with the distribution of distances, and the red
vertical line indicates the threshold value chosen to determine the short/long distance
criterion. A close distance between the variables indicates a robust statistical relationship.
Conversely, a high distance indicates no clear relationship between the variables and,
therefore, does not appear with any label (no color code). For example, the first product in
Figure 6.
The three main
eigenvectors (a) and
the representation of
the bootstrap
resamples together
with their confidence
volumes (b) from the
first experiment
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
(a)
(b)
Source: Author’s own creation
JHTT

the simplified table (“1/2 cazuelita callos”) has a strong relationship with February and a
medium relationship with January. Another general conclusion (with a panoramic view) is
the high sales of perishable products in February and October.
Based on Figure 5, it can be observed that perishable items tend to have longer distances
in January and August, whereas February, June and October tend to have smaller distances.
These distances represent the probability of concurrence between different product
categories. Short distances indicate a high probability of category concurrence, implying
that specific product categories have increased sales during those months. On the other
hand, longer distances indicate a lower probability of category concurrence, suggesting that
those product categories have dissimilar sales during those months.
Figure 7.
Histogram of
empirical sample
distances (a) and the
simplified table with
the compilation of
relationships (b) from
the first experiment
Avoiding food
waste

Considering these observations, specific business strategies can be developed to address
each scenario. In the case of short distances, where product categories have a high
probability of concurrence, inventory management and cross-promotion tactics can be
implemented. For example, if a restaurant offers both meats and wines, it could be beneficial
during February, June and October to create special offers that combine both products, as
there is a shared demand for them. On the other hand, in the case of longer distances where
the probability of concurrence is low, RM strategies can be applied to maximize profits. This
could involve dynamically adjusting prices based on the specific sale for each product
category, conducting profitability analysis and adjusting marketing strategies accordingly
or even exploring opportunities for product delivery management.
4.2 Experiment on product monthly seasonality
In this second experiment, we searched for relationships among all the perishable products
(i.e. seasonal and nonseasonal products) and the months of the year. Similarly to the
previous experiment, each row in the data matrix accounts for the product categories and
month for a given ticket. For instance, drawing upon the information in Figure 1, a product
value could be “Cantabrian hake”and the corresponding month value “March”. The data set
is different in this case since the 190 perishable products were used without restriction.
Many presented a residual value; therefore, the data set was filtered to 64 perishable
products. In Experiment 1, the perishable products are offered continuously throughout the
year, so with both experiments, we want to observe the influence of seasonality on consumer
behavior. Thus, our distance table dimension is 6412 (number of perishable products per
number of months).
In the MCA study, the results of the three principal eigenvectors are shown in Figure 8(a).
In addition to having significant categories in the first two, five significant categories are
obtained in the third projection (different from the previous experiment), which considerably
improves the visualization of the bootstrap resamples and their confidence volumes
(because the clouds are more compact in this case) as shown in Figure 8(b).More
differentiated relationships are observed (although fewer than in the previous experiment).
This difference from the previous experiment is due to the behavior of the unsupervised
MCA model, which groups the less strong relationships in the center and moves the stronger
ones away from the center.
Figure 9 shows the essential relationships following the same color criteria of the
previous experiment, but a new black color code is included. This new color code indicates
the products not consumed in these months to identify the noise. In the relationships
compiled in the simplified table, the products “ensalada tomate”,“secreto ib”and
“alcachofas fritas”stand out with a similar consumption trend (where the three coincide in
the months where there is no consumption), and thus presenting a high ratio in January,
February and March.
4.3 Recommendation system
A recommendation system based on collaborative filtering using historical consumption
data has been proposed. It is implemented in SwiftUI, making it compatible with iOS
devices. The app manages product stock using an unsupervised model, with products
labeled by color and capital letters indicating consumption ratios: “H”(red) for high
percentages, “M”(green) for medium and “N”(black) for products with no or zero recent
consumption. Using this, managers can adjust supplies based on historical consumption. As
of July, Figure 10 shows a demo of the app showcasing the labeling system from the second
experiment.
JHTT

5. Conclusions
Inefficiencies in demand and sales forecasting and subpar stock management result in food
wastage, underscoring the importance of accurate forecasting to prevent waste (Pirani and
Arafat, 2016;Young et al., 2018). Sales forecasting also helps short and long-term decision-
making, reducing costs and increasing sales (Doganis et al., 2006). Traditional restaurant
Figure 9.
The simplified table
with the compilation
of relationships from
the second
experiment
Figure 8.
The three main
eigenvectors (a) and
the representation of
the bootstrap
resamples together
with their confidence
volumes (b) from the
second experiment
-0.2
-0.1
0
0.1
0.2
0.3
0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
(a)
(b)
Source: Author’s own creation
Avoiding food
waste

forecasting has largely been intuitive, but its inherent complexities render judgmental
techniques to be updated. Modern restaurant management should harness technology for
more precise forecasting and dynamic business strategies (Filimonau and De Coteau, 2020).
Although restaurants predominantly rely on sales tickets for customer data, these tickets
reveal valuable insights for sales predictions. Using transactional data, restaurants can
enhance operational efficiency, product management and customer satisfaction through
dynamic pricing and menu engineering (Liu et al.,2001).
Forecasting and RM are closely linked, and both have a significant role in managing food
waste. RM systems use forecasting to reduce waste by optimizing inventory and improving
performance. Simultaneously, this efficient management of perishable resources minimizes
food waste. Therefore, effective RM management supported by accurate forecasts can serve
as a comprehensive approach to resource management, maximizing profits and minimizing
food waste (Weatherford and Kimes, 2003;Kimes et al., 1998).
This study exploits restaurant ticket data, pioneering using unsupervised models in the
food sector for detailed sales predictions. The integrated model streamlines product
forecasting by analyzing relationships between products, days and other variables, saving
managerial analysis time (Cadavid et al.,2018).
Figure 10.
Main screen of the
recommender system
application
JHTT

The key outcome of the conducted experiments is the ability to examine the connections
between products and data using a unified model. Using ticket restaurant information could
apply this approach to other variables, such as days, hours, tables and restaurant areas.
These tools eliminate the need for individually filtering each product, creating separate
models and predicting monthly sales for eachproduct to obtain monthly deals. By using this
approach, managers can save time and effort in analyzing and forecasting sales.
There are different results the manager should investigate, such as the similar
consumption of four products and the fact that three others were only consumed in a few
months. In the first case, these results could indicate that these are products that
customers order together (so it is possible to suggest their joint sale: cross-selling). In the
second case, they could have been newly introduced on the menu, or they should be
seasonal products. This provides consumption patterns for the manager to make a more
accurate forecast.
The bootstrap tool enhances our understanding of monthly product sales trends by
providing more statistically robust data than conventional frequency-based probabilities
typically derived from historical data analysis. This improved level of precision supports
managers in making informed decisions about inventory management (Dixon, 2006). In
addition, the SVDD method offers a visualization of confidence volumes, or clouds, that
encapsulate the relationships between perishable products and specific months. These
relationships, translated into table color codes, provide insights for managers about how
products relate to different periods. A more precise tool for determining category overlap
would further enhance this utility.
The process can illuminate connections between regularly sold products throughout
the year, offering valuable data for a restaurant manager’s planning. These statistics and
experiments would only be helpful if they could be easily applied. An application based
on a recommendation system has been proposed that allows the manager to plan
perishable products. The visualization of the products in the application has different
color-coded labels that show the relationships between the monthly sales habits of
perishable products.
This information will improve pricing, accurate sales forecasting and menu
engineering strategies. In addition, it helps to match product inventory and purchasing
with demand to avoid food waste and optimize restaurant profit. The unsupervised
model extracts the most important features (through dimensionality reduction) and
fulfills three of the seven functions attributed to BD: data visualization, velocity and
veracity. To the best of our knowledge, no previous work has focused on the challenge of
generating product recommendations in restaurants using unsupervised models,
particularly by leveraging restaurant ticket data.
6. Theoretical implications
Unsupervised models offer a deeper understanding of intricate data relationships,
addressing the challenge of black-box phenomena commonly linked with ML models
(Saunshi et al., 2019). This research taps into MCA to explore the relationships between
products and variables in restaurant tickets. While MCA is a linear technique, the study
highlights its potential to reveal operative relationships. With the flourishing field of
manifold learning, introducing nonlinear techniques may provide further insights into
such data (Martinez-Mateu et al.,2023;Chaquet-Ulldemolins et al., 2022).
On the other hand, it is worth highlighting the relevance of this RM study, especially
since COVID-19 has facilitated RM application via dynamic charts, with forecasting being a
key input for establishing RM strategies.
Avoiding food
waste

7. Practical implications
The primary practical implication of this research is developing a user-friendly recommendation
system for restaurant managers. This system uses visually appealing information to provide
insights into food and beverage consumption. Its user-friendly nature, coupled with the easy
accessibility of restaurant ticket data, enables its adoption by small and medium-sized
enterprises in the restaurant industry (Lee et al., 2018).
The impact of this research can be twofold. First, sales forecasting addresses
sustainability concerns by reducing food waste (Özbük and Coskun, 2020). Second, it
supports inventory management by aligning supply and sales, saving costs and
improving company profitability.
Additionally, the system facilitates input forecasting by accurately forecasting sales and
the future possibility of linking them to dish components. This information becomes
invaluable for making informed decisions regarding supplies. Moreover, the system can
provide relevant insights for sales forecasting by understanding customer consumption
patterns (e.g. correlating sales with different table types such as individuals, couples or
families). Precise sales forecasting leads to better price adjustments, capacity planning and
sales optimization,ultimately maximizing company profits (Lee et al.,2020).
8. Limitations and future research
The limited information on restaurant tickets poses a significant constraint. Predicting purchases
rather than sales is preferable, but this requires digitizing recipes with the exact weight of each
product, something yet to be commonplace in many restaurants. Unsupervised models can
generalize without prior training, but they have the drawback of varying implementation
behaviors. The MCA method reveals strong statistical relationships but is sensitive to sample
sizes. Moreover, the clustering of products in three-dimensional representations can obscure
transparent relationships. Changes introduced by menu managers might produce false positives
in the model, potentially affecting new products. An additional layer could be added to the
recommendation system to address this.
There are opportunities to further research on frequency relationships between
perishable products and temporal frequencies. On the other hand, studying the relationships
between products for cross-selling strategies is also possible. Given the limited ticket
information, products could be detailed further and linked with other databases, such as
weather conditions, to assess their impact on sales. While more advanced forecasting
models using ML can be developed, these would require additional data acquisition and
monitoring efforts. The unsupervised recommender system is a practical tool using
historical data but could pave the way for adopting more sophisticated forecasting models.
References
Bang, S., Bishnoi, R., Chauhan, A.S., Dixit, A.K. and Chawla, I. (2019), “Fuzzy logic based crop yield
prediction using temperature and rainfall parameters predicted through Arma, Sarima, and
armax models”,2019 Twelfth International Conference on Contemporary Computing (IC3),
IEEE, pp. 1-6.
Bote-Curiel, L., Munoz-Romero, S., Gerrero-Curieses, A. and Rojo-Álvarez, J.L. (2019), “Deep learning
and big data in healthcare: a double review for critical beginners”,Applied Sciences, Vol. 9 No. 11,
p. 2331, doi: 10.3390/app9112331.
Cadavid, J.P.U., Lamouri, S. and Grabot, B. (2018), “Trends in machine learning applied to demand and
sales forecasting: a review”,International Conference on Information Systems, Logistics and
Supply Chain.
JHTT

Carpenter, J. and Bithell, J. (2000), “Bootstrap confidence intervals: when, which, what? A practical
guide for medical statisticians”,Statistics in Medicine, Vol. 19 No. 9, pp. 1141-1164, doi: 10.1002/
(SICI)1097-0258(20000515)19:9<1141::AID-SIM479>3.0.CO;2-F.
Cavusoglu, M. (2019), “An analysis of technology applications in the restaurant industry”,Journal of
Hospitality and Tourism Technology, Vol. 10 No. 1, pp. 45-72, doi: 10.1108/JHTT-12-2017-0141.
Chaquet-Ulldemolins, J., Gimeno-Blanes, F.J., Moral-Rubio, S., Muñoz-Romero, S. and Rojo-Álvarez, J.L.
(2022), “On the black-box challenge for fraud detection using machine learning (ii): nonlinear
analysis through interpretable autoencoders”,Applied Sciences, Vol. 12 No. 8, pp. 1-16, doi:
10.3390/app1208.3856.
Corral-De-Witt, D., Carrera, E.V., Muñoz-Romero, S., Tepe, K. and Rojo-Álvarez, J.L. (2019), “Multiple
correspondence analysis of emergencies attended by integrated security services”,Applied
Sciences, Vol. 9 No. 7, p. 1396.
Cunningham, P., Cord, M. and Delany, S.J. (2008), Supervised Learning, Machine Learning Techniques
for Multimedia, Springer, Berlin, pp. 21-49.
Derqui, B., Fernandez, V. and Fayos, T. (2018), “Towards more sustainable food systems. addressing
food waste at school canteens”,Appetite, Vol. 129, pp. 1-11.
Dhir, A., Talwar, S., Kaur, P. and Malibari, A. (2020), “Food waste in hospitality and food services: a
systematic literature review and framework development approach”,Journal of Cleaner
Production, Vol. 270, p. 122861.
Dixon, P.M. (2006), “Bootstrap resampling”,Encyclopedia of Environmetrics, Wiley, London.
do Carmo Stangherlin, I. and De Barcellos, M.D. (2018), “Drivers and barriers to food waste reduction”,
British Food Journal, Vol. 120 No. 10.
Doborjeh, Z., Hemmington, N., Doborjeh, M. and Kasabov, N. (2022), “Artificial intelligence: a
systematic reviewof methods and applications in hospitality and tourism”,International Journal
of Contemporary Hospitality Management, Vol. 34 No. 3, pp. 1154-1176.
Doganis, P., Alexandridis, A., Patrinos, P. and Sarimveis, H. (2006), “Time series sales forecasting for
short shelf-life food products based on artificial neural networks and evolutionary computing”,
Journal of Food Engineering, Vol. 75 No. 2, pp. 196-204.
Dursun, A. and Caber, M. (2016), “Using data mining techniques for profiling profitable hotel customers:
an application of RFM analysis”,Tourism Management Perspectives, Vol. 18, pp. 153-160.
Engström, R. and Carlsson-Kanyama, A. (2004), “Food losses in food service institutions examples from
Sweden”,Food Policy, Vol. 29 No. 3, pp. 203-213.
Filimonau, V. and De Coteau, D. (2020), “Tourism resilience in the context of integrated destination
and disaster management (dm2)”,International Journal of Tourism Research, Vol. 22 No. 2,
pp. 202-222, doi: 10.1002/jtr.2329.
Ghosh, P.R., Fawcett, S.B., Sharma, G. and Poinern, E.J. (2016), “Progress towards sustainable
utilisation and management of food wastes in the global economy”,International Journal of
Food Science, Vol. 2016.
Glynn, D. (2014), “Correspondence analysis”,Methods for Semantics: Quantitative Studies in Polysemy
and Synonymy, John Benjamins Publishing, Amsterdam, pp. 443-486.
Gonz
alez-Serrano, L. and Tal
on-Ballestero, P. (2020), “Revenue management and e-tourism: the past,
present and future”,Handbook of e-Tourism, Springer, Cham, pp. 1-28.
Gonz
alez-Serrano, L., Tal
on-Ballestero, P., Muñoz-Romero, S., Soguero-Ruiz, C. and Rojo-Álvarez, J.L. (2020),
“A big data approach to customer relationship management strategy in hospitality using multiple
correspondence domain description”,Applied Sciences, Vol. 11 No. 1, p. 256, doi: 10.3390/app11010256.
Goonan, S., Mirosa, M. and Spence, H. (2014), “Getting a taste for food waste: a mixed methods
ethnographic study into hospital food waste before patient consumption conducted at three New
Zealand foodservice facilities”,Journal of the Academy of Nutrition and Dietetics, Vol. 114 No. 1,
pp. 63-71.
Avoiding food
waste

Gunders, D. and Bloom, J. (2017), “Wasted: How America is losing up to 40% of its food from farm to
fork to landfill”, Natural Resources Defense Council New York.
Hall, P. and Martin, M.A. (1988), “On bootstrap resampling and iteration”,Biometrika, Vol. 75 No. 4,
pp. 661-671, doi: 10.1093/biomet/75.4.661.
Hennchen, B. (2019), “Knowing the kitchen: applying practice theory to issues of food waste in the food
service sector”,Journal of Cleaner Production, Vol. 225, pp. 675-683.
Hsu, H. (2021), “Book review ‘big data and innovation in tourism, travel, and hospitality: managerial
approaches, techniques, and applications”.
Hsu, M.J., Ting, H., Lui, T.W., Chen, S.C. and Cheah, J.H. (2022), “Guest editorial: challenges and
prospects of AIoT application in hospitality and tourism marketing”,Journal of Hospitality and
Tourism Technology, Vol. 13 No. 3, pp. 349-355.
Husson, F. and Josse, J. (2014), “Multiple correspondence analysis”,Visualization and Verbalization of
Data, The CRC Press, Boca Raton, pp. 165-184.
Janikow, C.Z. (1993), “A knowledge-intensive genetic algorithm for supervised learning”,Genetic
Algorithms for Machine Learning, Springer, Cham, pp. 33-72.
Kim, D.J. and Lee, S. (2019), “A study of consumer perception on fashion show using big data analysis”,
Journal of Fashion Business, Vol. 23 No. 3, pp. 85-100.
Kimes, S.E., Chase, R.B., Choi, S., Lee, P.Y. and Ngonzi, E.N. (1998), “Restaurant revenue management:
applying yield management to the restaurant industry”,Cornell Hotel and Restaurant
Administration Quarterly, Vol. 39 No. 3, pp. 32-39.
Lasek, A., Cercone, N. and Saunders, J. (2016), “Smart restaurants: survey on customer demand and
sales forecasting”,Smart Cities and Homes, Elsevier, Cham, pp. 361-386.
Lee, W.H., Lin, C.W. and Shih, K.H. (2018), “A technology acceptance model for the perception of
restaurant service robots for trust, interactivity, and output quality”,International Journal of
Mobile Communications, Vol. 16 No. 4, pp. 361-376.
Lee, M., Mu, X. and Zhang, Y. (2020), “A machine learning approach to improving forecasting accuracy
of hotel demand: a comparative analysis of neural networks and traditional models”,Issues in
Information Systems, Vol. 21 No. 1.
Liu, L.M., Bhattacharyya, S., Sclove,S.L., Chen, R. and Lattyak, W.J. (2001), “Data mining on time series:
an illustration using fast-food restaurant franchise data”,Computational Statistics and Data
Analysis, Vol. 37 No. 4, pp.455-476.
Liu, Y.H., Lin, S.H., Hsueh, Y.L. and Lee, M.J. (2009), “Automatic target defect identification for TFT-
LCD array process inspection using kernel FCM-based fuzzy SVDD ensemble”,Expert Systems
with Applications, Vol. 36 No. 2, pp. 1978-1998.
McCray, S., Maunder, K., Krikowa, R. and MacKenzie-Shalders, K. (2018), “Room service
improves nutritional intake and increases patient satisfaction while decreasing food
waste and cost”,Journal of the Academy of Nutrition and Dietetics, Vol. 118 No. 2,
pp. 284-293.
Mariani, M. (2019), “Big data and analytics in tourism and hospitality: a perspective article”,Tourism
Review, Vol. 75 No. 1, pp. 299-303, doi: 10.1108/TR-06-2019-0259.
Mariani, M., Baggio, R., Fuchs, M. and Höepken, W. (2018), “Business intelligence and big data in
hospitality and tourism: a systematic literature review”,International Journal of Contemporary
Hospitality Management, Vol. 30 No. 12, pp. 3514-3554.
Martinez-Mateu, L., Melgarejo-Meseguer, F.M., Muñoz-Romero, S., Gimeno-Blanes, F.J., García-
Alberola, A., Rocher-Ventura, S., Saiz, J. and Rojo-Álvarez, J.L. (2023), “Manifold analysis of the
p-wave changes induced by pulmonary vein isolation during cryoballoon procedure”,
Computers in Biology and Medicine, Vol. 155, p. 106655.
Martin-Rios, C., Demen-Meier, C., Gössling, S. and Cornuz, C. (2018), “Food waste management innovations
in the foodservice industry”,Waste Management, Vol. 79, pp. 196-206.
JHTT

Meek, S., Wilk, V. and Lambert, C. (2021), “A big data exploration of the informational and normative
influences on the helpfulness of online restaurant reviews”,Journal of Business Research,
Vol. 125, pp. 354-367.
Mirosa, M., Liu, Y. and Mirosa, R. (2018), “Consumers’behaviors and attitudes toward doggy bags:
identifying barriers and benefits to promoting behavior change”,Journal of Food Products
Marketing, Vol. 24 No. 5, pp. 563-590.
Mond
ejar-Jim
enez, J.A., Ferrari, G., Secondi, L. and Principato, L. (2016), “From the table to waste: an
exploratory study on behaviour towards food waste of Spanish and Italian youths”,Journal of
Cleaner Production, Vol. 138, pp. 8-18.
Murakami, T. (2020), “Orthonormal principal component analysis for categorical data as a
transformation of multiple correspondence analysis”,Advanced Studies in Behaviormetrics and
Data Science, Springer, Cham, pp. 211-231.
Nadkarni, S., Kriechbaumer, F., Rothenberger, M. and Christodoulidou, N. (2019), “The path to the hotel
of things: Internet of things and big data converging in hospitality”,Journal of Hospitality and
Tourism Technology, Vol. 11 No. 1.
Ogbeide, G.C., Fu, Y.Y. and Cecil, A.K. (2020), “Are hospitality/tourism curricula ready for big data?”,
Journal of Hospitality and Tourism Technology,Vol.12No.1,doi:10.1108/JHTT-09-2017-0081.
Özbük, R.M.Y. and Coskun, A. (2020), “Factors affecting food waste at the downstream entities of the
supply chain: a critical review”,Journal of Cleaner Production, Vol. 244, p. 118628.
Pinto, R.S., dos Santos Pinto, F.F.S., Melo, S.S., Campos, C. and Cordovil, M. (2018), “A simple
awareness campaign to promote food waste reduction in a university canteen”,Waste
Management, Vol. 76, pp. 28-38.
Pirani, S.I. and Arafat, H.A. (2016), “Reduction of food waste generation in the hospitality industry”,
Journal of Cleaner Production, Vol. 132, pp. 129-145.
Posch, K., Truden, C., Hungerländer, P. and Pilz, J. (2022), “A Bayesian approach for predicting food
and beverage sales in staff canteens and restaurants”,International Journal of Forecasting,
Vol. 38 No. 1, pp. 321-338.
Priyadarshi, N., Padmanaban, S., Holm-Nielsen, J.B., Blaabjerg, F. and Bhaskar, M.S. (2019), “An
experimental estimation of hybrid ANFIS–PSO-based MPPT for PV grid integration under
fluctuating sun irradiance”,IEEE Systems Journal, Vol. 14 No. 1, pp. 1218-1229.
Rizou, M., Galanakis, I.M., Aldawoud, T.M. and Galanakis, C.M. (2020), “Safety of foods, food supply
chain and environment within the COVID-19 pandemic”,Trends in Food Science and
Technology, Vol. 102, pp. 293-299.
Sakib, S.N. (2023), “Restaurant sales prediction using machine learning”,Handbook of Research on AI and
Machine Learning Applications in Customer Support and Analytics, IGI Global, Hershey, pp. 202-226.
Samara, D., Magnisalis, I. and Peristeras, V. (2020), “Artificial intelligence and big data in tourism: a
systematic literature review”,Journal of Hospitality and Tourism Technology, Vol. 11 No. 2,
pp. 343-367, doi: 10.1108/JHTT-12-2018-0118.
Saunshi, N., Plevrakis, O., Arora, S., Khodak, M. and Khandeparkar, H. (2019), “A theoretical analysis of
contrastive unsupervised representation learning”, International Conference on Machine
Learning, PMLR, pp. 628-5637.
Schmidt, A., Kabir, M.W.U. and Hoque, M.T. (2022), “Machine learning based restaurant sales forecasting”,
Machine Learning and Knowledge Extraction, Vol. 4 No. 1, pp. 105-130, doi: 10.3390/make4010006.
Shilo, S., Rossman, H. and Segal, E. (2020),“Axes of a revolution: challenges and promises of big data in
healthcare”,Nature Medicine, Vol. 26 No. 1, pp. 29-38.
Soguero-Ruiz, C., Alonso-Arteaga, N., Muñoz-Romero, S., Rojo-Álvarez, J.L., Rubio-S
anchez, M., L
opez-
Fajardo, I.C. and Mora-Jim
enze, I. (2020), “Finding associations among chronic conditions by
bootstrap and multiple correspondence analysis”,2020 IEEE International Conference on
Bioinformatics and Biomedicine (BIBM), IEEE, pp. 2066-2073.
Avoiding food
waste

Song, H., Qiu, R.T. and Park, J. (2019), “A review of research on tourism demand forecasting: launching
the annals of tourism research curated collection on tourism demand forecasting”,Annals of
Tourism Research, Vol. 75, pp. 338-362.
Srijuntrapun, P., Sukwong, P. and Marshall, A. (2022), “The role of food waste hierarchy as Thai hotels
seek to fulfill their corporate social responsibility”,Heliyon, Vol. 8 No. 10, p. e11201.
Tanizaki, T., Hoshino, T., Shimmura, T. and Takenaka, T. (2019), “Demand forecasting in restaurants
using machine learning and statistical analysis”,Procedia CIRP, Vol. 79, pp. 679-683.
Tao, Z., Wang, D., Yao, F., Huang, X., Wu, Y., Du, M., Chen, Z., An, H., Li, X. and Yang, Q. (2020), “The
effects of thiosulfinates on methane production from anaerobic co-digestion of waste activated
sludge and food waste and mitigate method”,Journal of Hazardous Materials, Vol. 384, p. 121363.
Tax, D.M. and Duin, R.P. (2004), “Support vector data description”,Machine Learning, Vol. 54 No. 1,
pp. 45-66.
Terpstra, M., Steenbekkers, L., De Maertelaere, N. and Nijhuis, S. (2005), “Food storage and disposal:
consumer practices and knowledge”,British Food Journal, Vol. 107 No. 7.
Tsoumakas, G. (2019), “A survey of machine learning techniques for food sales prediction”,Artificial
Intelligence Review, Vol. 52 No. 1, pp. 441-447.
van der Vorst, J.G., Beulens, A.J., De Wit, W. and van Beek, P. (1998), “Supply chain management in
food chains: improving performance by reducing uncertainty”,International Transactions in
Operational Research, Vol. 5 No. 6, pp. 487-499, doi: 10.1111/j.1475-3995.1998.tb00131.x.
van Leeuwen, R. and Koole, G. (2022), “Demand forecasting in hospitality using smoothed demand
curves”,Journal of Revenue and Pricing Management, Vol. 21 No. 5, pp. 487-502, doi: 10.1057/
s41272-021-00364-5.
Vargas-Calder
on, V., Moros Ochoa, A., Castro Nieto, G.Y. and Camargo, J.E. (2021), “Machine learning
for assessing quality of service in the hospitality sector based on customer reviews”,
Information Technology and Tourism, Vol. 23 No.3, pp. 351-379.
Weatherford, L.R. and Kimes, S.E. (2003), “A comparison of forecasting methods for hotel revenue
management”,International Journal of Forecasting, Vol. 19 No. 3, pp. 401-415.
Wood, M. (2005), “Bootstrapped confidence intervals as an approach to statistical inference”,
Organizational Research Methods, Vol. 8 No. 4, pp. 454-470, doi: 10.1177/1094428105280059.
Young, C.W., Russell, S.V., Robinson, C.A. and Chintakayala, P.K. (2018), “Sustainable retailing–
influencing consumer behaviour on food waste”,Business Strategy and the Environment, Vol. 27
No. 1, pp. 1-15.
Corresponding author
Ismael G
omez-Talal can be contacted at: ismael.gomez.talal@urjc.es
For instructions on how to order reprints of this article, please visit our website:
www.emeraldgrouppublishing.com/licensing/reprints.htm
Or contact us for further details: permissions@emeraldinsight.com
JHTT