No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Time Series Trend Detection and Forecasting Using Complex Network Topology Analysis
... The algorithm, which first converts the initial time series to a fuzzy time series, was used to forecast the operation of the automated evaporating station. The next step is to convert the obtained fuzzy time series into a time series of fuzzy elementary tendencies and to perform defuzzification using the method of the center of gravity of intensity of each fuzzy elementary tendency for each time series () tt a DeFuzzy a (Anghinoni et al., 2018). The analysis of the stability of the prediction model is as follows. ...
... Let's calculate the amount of the intensities of fuzzy elementary tendencies for each interval by the following way (Dong et al., 2017): With the developed algorithm, local tendencies are assessed. The next step is to use language and numerical forms in the algorithm (Anghinoni et al., 2018). For the operation of this algorithm it is necessary to convert the initial time series into a fuzzy time series (Mehmood et al., 2020) using the model shown in Figure 2. The next step in this algorithm is to divide the obtained time series into a number of intervals. ...
... This algorithm has a disadvantage due to the limitation of its operation by the number of predefined time intervals. Therefore, the number of identified local tendencies will be equal to the number of intervals specified by the developer (Anghinoni et al., 2018). This algorithm allows obtaining time series that can be used in the future to forecast local tendencies. ...
... The algorithm, which first converts the initial time series to a fuzzy time series, was used to forecast the operation of the automated evaporating station. The next step is to convert the obtained fuzzy time series into a time series of fuzzy elementary tendencies and to perform defuzzification using the method of the center of gravity of intensity of each fuzzy elementary tendency for each time series ( ) t t a DeFuzzy a (Anghinoni et al., 2018). ...
... With the developed algorithm, local tendencies are assessed. The next step is to use language and numerical forms in the algorithm (Anghinoni et al., 2018). For the operation of this algorithm it is necessary to convert the initial time series into a fuzzy time series (Mehmood et al., 2020) using the model shown in Figure 2. The next step in this algorithm is to divide the obtained time series into a number of intervals. ...
... This algorithm has a disadvantage due to the limitation of its operation by the number of predefined time intervals. Therefore, the number of identified local tendencies will be equal to the number of intervals specified by the developer (Anghinoni et al., 2018). This algorithm allows obtaining time series that can be used in the future to forecast local tendencies. ...
... Time series data are omnipresent in many practical data science applications ranging from health care (Gao et al., 2014) and stock market predictions (Anghinoni et al., 2018) to social media analysis (Xu, Chen, and Mao, 2018) and human activity recognition (Xi et al., 2018). In fact, any type of numerical acquisition of data with some notion of ordering will generate time series, making this type of data very common among data mining problems (Längkvist, Karlsson, and Loutfi, 2014). ...
... A concrete example of time series data in health care would be the acquisition of ECG heart signals (Rajan and Thiagarajan, 2018). In stock market analysis, a time series element would correspond to the value of a stock at a given time stamp (Anghinoni et al., 2018). In human activity recognition, the given Cartesian position of a hand in 3D space would constitute an element of a time series (Ignatov, 2018). ...
Data science is about designing algorithms and pipelines for extracting knowledge from large masses of data.Time series analysis is a field of data science which is interested in analyzing sequences of numerical values ordered in time.Time series are particularly interesting because they allow us to visualize and understand the evolution of a process over time.Their analysis can reveal trends, relationships and similarities across the data.There exists numerous fields containing data in the form of time series: health care (electrocardiogram, blood sugar, etc.), activity recognition, remote sensing, finance (stock market price), industry (sensors), etc.In data mining, classification is a supervised task that involves learning a model from labeled data organized into classes in order to predict the correct label of a new instance.Time series classification consists of constructing algorithms dedicated to automatically label time series data.For example, using a labeled set of electrocardiograms from healthy patients or patients with a heart disease, the goal is to train a model capable of predicting whether or not a new electrocardiogram contains a pathology.The sequential aspect of time series data requires the development of algorithms that are able to harness this temporal property, thus making the existing off-the-shelf machine learning models for traditional tabular data suboptimal for solving the underlying task.In this context, deep learning has emerged in recent years as one of the most effective methods for tackling the supervised classification task, particularly in the field of computer vision.The main objective of this thesis was to study and develop deep neural networks specifically constructed for the classification of time series data.We thus carried out the first large scale experimental study allowing us to compare the existing deep methods and to position them compared other non-deep learning based state-of-the-art methods.Subsequently, we made numerous contributions in this area, notably in the context of transfer learning, data augmentation, ensembling and adversarial attacks.Finally, we have also proposed a novel architecture, based on the famous Inception network (Google), which ranks among the most efficient to date.Our experiments carried out on benchmarks comprising more than a hundred data sets enabled us to validate the performance of our contributions.Finally, we also showed the relevance of deep learning approaches in the field of surgical data science where we proposed an interpretable approach in order to assess surgical skills from kinematic multivariate time series data.
... Time series data are omnipresent in many practical data science applications ranging from health care (Gao et al., 2014) and stock market predictions (Anghinoni et al., 2018) to social media analysis (Xu, Chen, and Mao, 2018) and human activity recognition (Xi et al., 2018). In fact, any type of numerical acquisition of data with some notion of ordering will generate time series, making this type of data very common among data mining problems (Längkvist, Karlsson, and Loutfi, 2014). ...
... A concrete example of time series data in health care would be the acquisition of ECG heart signals (Rajan and Thiagarajan, 2018). In stock market analysis, a time series element would correspond to the value of a stock at a given time stamp (Anghinoni et al., 2018). In human activity recognition, the given Cartesian position of a hand in 3D space would constitute an element of a time series (Ignatov, 2018). ...
Time series analysis is a field of data science which is interested in analyzing sequences of numerical values ordered in time. Time series are particularly interesting because they allow us to visualize and understand the evolution of a process over time. Their analysis can reveal trends, relationships and similarities across the data. There exists numerous fields containing data in the form of time series: health care (electrocardiogram, blood sugar, etc.), activity recognition, remote sensing, finance (stock market price), industry (sensors), etc. Time series classification consists of constructing algorithms dedicated to automatically label time series data. The sequential aspect of time series data requires the development of algorithms that are able to harness this temporal property, thus making the existing off-the-shelf machine learning models for traditional tabular data suboptimal for solving the underlying task. In this context, deep learning has emerged in recent years as one of the most effective methods for tackling the supervised classification task, particularly in the field of computer vision. The main objective of this thesis was to study and develop deep neural networks specifically constructed for the classification of time series data. We thus carried out the first large scale experimental study allowing us to compare the existing deep methods and to position them compared other non-deep learning based state-of-the-art methods. Subsequently, we made numerous contributions in this area, notably in the context of transfer learning, data augmentation, ensembling and adversarial attacks. Finally, we have also proposed a novel architecture, based on the famous Inception network (Google), which ranks among the most efficient to date.
... The research conducted in the realm of scientific and practical advancements concerning the formation of intricate signal ensembles distinguished by temporal separation through the employment of a neural network training system has yielded compelling insights. It reveals a unique dichotomy wherein the explorations, methodologies, and scientific innovations within this domain find partial application among foreign scholars [1][2][3][4][5], whereas Ukrainian researchers [6][7] approach the matter from a distinctive vantage point. The Ukrainian perspective is anchored in the conceptualization of the challenge as the creation of ensembles that encompass intricate signal-code structures woven into the tapestry of telecommunications signals, all within the expansive framework of telecommunications cognitive systems. ...
In telecommunications, time-division signals play an extremely important role in the process of information transmission and processing. These signals, which are presented as electromagnetic waves, serve as the main means of transport for the transmission of data from one communication device to another, effectively constituting the circulatory system of modern communication networks. The defining feature that distinguishes these signals is their temporal nature, or more precisely, the changes they undergo over time as they travel over the network. These signals contain a rich spectrum of parameters, covering not only the attributes of amplitude, frequency and phase, but also many other parameters. This article provides a scientific rationale for the feasibility and relevance of researching the process of forming signal ensembles with temporal separation using neural networks. It has been established that one of the key tasks in creating ensembles of complex signals with temporal separation through the application of neural networks is the selection of an optimal neural network architecture. This entails determining the type of neural network that best aligns with a specific research objective. An algorithm for forming signal ensembles with temporal separation using a recurrent neural network (RNN) has been developed. Special attention is devoted to methods for assessing the overall error of signal ensembles with temporal separation using PNN. These analytical methods encompass the utilization of pertinent metrics, the evaluation of which depends on the specific nature of the task, whether it involves regression or classification. The findings of this research will contribute to the further advancement of methods for forming signal ensembles with temporal separation through the utilization of neural networks and enhance their application in contemporary telecommunications systems.
... The spectrum of used AI techniques in finance field is wide and it includes since reinforcement Learning [7], [8], multiagent systems [9], [10] complex networks [11], decision trees [12], genetic algorithms [13], random forests [14] to more recent approaches like convolutional neural networks [15] and deep reinforcement learning [16]. Regardless of the picked AI technology, there are some aspects that are always present. ...
Autonomous trading robots have been studied in artificial intelligence area for quite some time. Many AI techniques have been tested for building autonomous agents able to trade financial assets. These initiatives include traditional neural networks, fuzzy logic, reinforcement learning but also more recent approaches like deep neural networks and deep reinforcement learning. Many developers claim to be successful in creating robots with great performance when simulating execution with historical price series, so called backtesting. However, when these robots are used in real markets frequently they present poor performance in terms of risks and return. In this paper, we propose an open source framework (mt5se) that helps the development, backtesting, live testing and real operation of autonomous traders. We built and tested several traders using mt5se. The results indicate that it may help the development of better traders. Furthermore, we discuss the simple architecture that is used in many studies and propose an alternative multiagent architecture. Such architecture separates two main concerns for portfolio manager (PM) : price prediction and capital allocation. More than achieve a high accuracy, a PM should increase profits when it is right and reduce loss when it is wrong. Furthermore, price prediction is highly dependent of asset's nature and history, while capital allocation is dependent only on analyst's prediction performance and assets' correlation. Finally, we discuss some promising technologies in the area
... The spectrum of used AI techniques in finance field is wide and it includes since reinforcement Learning [2], [3], multiagent systems [4], [5] complex networks [6], decision trees [7], genetic algorithms [8], random forests [9] to more recent approaches like convolutional neural networks [10] and deep reinforcement learning [11]. There are many cases, where the developers are successful in creating strategies with great performance when executing with historical price series (so called backtesting). ...
There are many practitioners that create software to buy and sell financial assets in an autonomous way. There are some digital platforms that allow the development, test and deployment of trading agents (or robots) in simulated or real markets. Some of these work focus on very short horizons of investment, while others deal with longer periods. The spectrum of used AI techniques in finance field is wide. There are many cases, where the developers are successful in creating robots with great performance in historical price series (so called backtesting). Furthermore, some platforms make available thousands of robots that [allegedly] are able to be profitable in real markets. These strategies may be created with some simple idea or using complex machine learning schemes. Nevertheless, when they are used in real markets or with data not used in their training or evaluation frequently they present very poor performance. In this paper, we propose a method for testing Foreign Exchange (FX) trading strategies that can provide realistic expectations about strategy's performance. This method addresses many pitfalls that can fool even experience practitioners and researchers. We present the results of applying such method in several famous autonomous strategies in many different financial assets. Analyzing these results, we can realize that it is very hard to build a reliable strategy and many published strategies are far from being reliable vehicles of investment. These facts can be maliciously used by those who try to sell such robots, by advertising such great (and non repetitive) results, while hiding the bad but meaningful results. The proposed method can be used to select among potential robots, establishes minimal periods and requirements for the test executions. In this way, the method helps to tell if you really have a great trading strategy or you are just fooling yourself.
... Time series data are omnipresent in many practical data science applications ranging from health care [1] and stock market predictions [2] to social media analysis [3] and human activity recognition [4]. Since 2006, time series analysis has been considered one of the most challenging problems in data mining [5], and in a more recent poll it has been shown that 48% of data expert had analyzed time series data during their career, ahead of text and images [6]. ...
... Time series data are omnipresent in many practical data science applications ranging from health care [1] and stock market predictions [2] to social media analysis [3] and human activity recognition [4]. Since 2006, time series analysis has been considered one of the most challenging problems in data mining [5], and in a more recent poll it has been shown that 48% of data expert had analyzed time series data during their career, ahead of text and images [6]. ...
Deep neural networks have revolutionized many fields such as computer vision and natural language processing. Inspired by this recent success, deep learning started to show promising results for Time Series Classification (TSC). However, neural networks are still behind the state-of-the-art TSC algorithms, that are currently composed of ensembles of 37 non deep learning based classifiers. We attribute this gap in performance due to the lack of neural network ensembles for TSC. Therefore in this paper, we show how an ensemble of 60 deep learning models can significantly improve upon the current state-of-the-art performance of neural networks for TSC, when evaluated over the UCR/UEA archive: the largest publicly available benchmark for time series analysis. Finally, we show how our proposed Neural Network Ensemble (NNE) is the first time series classifier to outperform COTE while reaching similar performance to the current state-of-the-art ensemble HIVE-COTE.
At present, mainstream research on complex network evolution involves measuring the developing characteristics of network structure at the macro level. Few studies have focused on the internal mechanism of network structure change and the prediction of network structure evolution at the micro‐level. However, for some special complex network time series, the traditional analysis and research methods at the macro level are invalid, and the micro‐level analysis method needs to be adopted. Based on the evolution characteristics of nodes and edges in complex networks, this paper constructs a time series network analysis and prediction algorithm to explore the structural evolution trends of complex network time series and make a prediction. Based on the enterprise supply–demand relationship data, this paper constructs the enterprise supply–demand network time series. Then, we use the method proposed in this paper to analyse and predict the complex network time series and verify the rationality and feasibility of the proposed method.
Complex networks are a unified form of rep-resenting complex systems. Through this representation is possible to study dynamical systems and make time series analysis using network techniques. One common charac-teristic of many real world time series is the periodicity. Detecting these periods is interesting because it permits to forecast the series behavior. Some techniques have been proposed to search for these periods but many of them fail to deal with noisy data. In this paper, we present an algorithm for periodicity detection in noisy data based on community detection. First, the method transforms a time series by dividing it into intervals that are represented by vertices. Then, we apply community detection in order to cluster highly connected ranges. These clusters of vertices represent periodic changes in the series and can be used to detect periodicity. The efficiency of the proposed method is illustrated in a meteorological case study where we detected periodicity in a noisy temperature data.
This paper proposes a stochastic wind power model based on an autoregressive integrated moving average (ARIMA) process. The model takes into account the nonstationarity and physical limits of stochastic wind power generation. The model is constructed based on wind power measurement of one year from the Nysted offshore wind farm in Denmark. The proposed limited-ARIMA (LARIMA) model introduces a limiter and characterizes the stochastic wind power generation by mean level, temporal correlation and driving noise. The model is validated against the measurement in terms of temporal correlation and probability distribution. The LARIMA model outperforms a first-order transition matrix based discrete Markov model in terms of temporal correlation, probability distribution and model parameter number. The proposed LARIMA model is further extended to include the monthly variation of the stochastic wind power generation.
Complex networks are an important paradigm of modern complex systems sciences which allows quantitatively assessing the structural properties of systems composed of different interacting entities. During the last years, intensive efforts have been spent on applying network-based concepts also for the analysis of dynamically relevant higher-order statistical properties of time series. Notably, many corresponding approaches are closely related to the concept of recurrence in phase space. In this paper, we review recent methodological advances in time series analysis based on complex networks, with a special emphasis on methods founded on recurrence plots. The potentials and limitations of the individual methods are discussed and illustrated for paradigmatic examples of dynamical systems as well as for real-world time series. Complex network measures are shown to provide information about structural features of dynamical systems that are complementary to those characterized by other methods of time series analysis and, hence, substantially enrich the knowledge gathered from other existing (linear as well as nonlinear) approaches.
Time series databases consist of sequences of values that are calculated or retrieved at regular intervals of time. The values or the events are measured at equal intervals of time (hourly, weekly, yearly etc). Time series data is very large in size having high dimensionality and is updated at regular intervals of time. Mining of the time series data is considered as an important analysis approach in large number of application areas like medicine, fraud detection, stock market etc. There has been a huge amount of research going in the field of time series data mining.
This article proposes a model-based procedure to decompose a time series uniquely into mutually independent additive seasonal, trend, and irregular noise components. The series is assumed to follow the Gaussian ARIMA model. Properties of the procedure are discussed and an actual example is given.
Many complex systems in nature and society can be described in terms of networks capturing the intricate web of connections among the units they are made of1, 2, 3, 4. A key question is how to interpret the global organization of such networks as the coexistence of their structural subunits (communities) associated with more highly interconnected parts. Identifying these a priori unknown building blocks (such as functionally related proteins5, 6, industrial sectors7 and groups of people8, 9) is crucial to the understanding of the structural and functional properties of networks. The existing deterministic methods used for large networks find separated communities, whereas most of the actual networks are made of highly overlapping cohesive groups of nodes. Here we introduce an approach to analysing the main statistical features of the interwoven sets of overlapping communities that makes a step towards uncovering the modular structure of complex systems. After defining a set of new characteristic quantities for the statistics of communities, we apply an efficient technique for exploring overlapping communities on a large scale. We find that overlaps are significant, and the distributions we introduce reveal universal features of networks. Our studies of collaboration, word-association and protein interaction graphs show that the web of communities has non-trivial correlations and specific scaling properties.
Autoregressive integrated moving average (ARIMA) is one of the popular linear models in time series forecasting during the past three decades. Recent research activities in forecasting with artificial neural networks (ANNs) suggest that ANNs can be a promising alternative to the traditional linear methods. ARIMA models and ANNs are often compared with mixed conclusions in terms of the superiority in forecasting performance. In this paper, a hybrid methodology that combines both ARIMA and ANN models is proposed to take advantage of the unique strength of ARIMA and ANN models in linear and nonlinear modeling. Experimental results with real data sets indicate that the combined model can be an effective way to improve forecasting accuracy achieved by either of the models used separately.
Considering the time-series ARIMA(p, d, q) model and fuzzy regression model, this paper develops a fuzzy ARIMA (FARIMA) model and applies it to forecasting the exchange rate of NT dollars to US dollars. This model includes interval models with interval parameters and the possibility distribution of future values is provided by FARIMA. This model makes it possible for decision makers to forecast the best- and worst-possible situations based on fewer observations than the ARIMA model.
Traditionally, the autoregressive integrated moving average (ARIMA) model has been one of the most widely used linear models in time series forecasting. However, the ARIMA model cannot easily capture the nonlinear patterns. Support vector machines (SVMs), a novel neural network technique, have been successfully applied in solving nonlinear regression estimation problems. Therefore, this investigation proposes a hybrid methodology that exploits the unique strength of the ARIMA model and the SVMs model in forecasting stock prices problems. Real data sets of stock prices were used to examine the forecasting accuracy of the proposed model. The results of computational tests are very promising.
In many real situations, randomness is considered to be uncertainty or even confusion which impedes human beings from making a correct decision. Here we study the combined role of randomness and determinism in particle dynamics for complex network community detection. In the proposed model, particles walk in the network and compete with each other in such a way that each of them tries to possess as many nodes as possible. Moreover, we introduce a rule to adjust the level of randomness of particle walking in the network, and we have found that a portion of randomness can largely improve the community detection rate. Computer simulations show that the model has good community detection performance and at the same time presents low computational complexity.
The discovery and analysis of community structure in networks is a topic of considerable recent interest within the physics community, but most methods proposed so far are unsuitable for very large networks because of their computational cost. Here we present a hierarchical agglomeration algorithm for detecting community structure which is faster than many competing algorithms: its running time on a network with n vertices and m edges is O (md log n) where d is the depth of the dendrogram describing the community structure. Many real-world networks are sparse and hierarchical, with m approximately n and d approximately log n, in which case our algorithm runs in essentially linear time, O (n log(2) n). As an example of the application of this algorithm we use it to analyze a network of items for sale on the web site of a large on-line retailer, items in the network being linked if they are frequently purchased by the same buyer. The network has more than 400 000 vertices and 2 x 10(6) edges. We show that our algorithm can extract meaningful communities from this network, revealing large-scale patterns present in the purchasing habits of customers.
We propose and study a set of algorithms for discovering community structure in networks-natural divisions of network nodes into densely connected subgroups. Our algorithms all share two definitive features: first, they involve iterative removal of edges from the network to split it into communities, the edges removed being identified using any one of a number of possible "betweenness" measures, and second, these measures are, crucially, recalculated after each removal. We also propose a measure for the strength of the community structure found by our algorithms, which gives us an objective metric for choosing the number of communities into which a network should be divided. We demonstrate that our algorithms are highly effective at discovering community structure in both computer-generated and real-world network data, and show how they can be used to shed light on the sometimes dauntingly complex structure of networked systems.