Résumé

Stock markets have an important impact on economic growth of countries. The prediction
of stock market indexes has been a complex task for last years. Indeed, many researches and financial analysts are highly interested in the research area of stock market prediction. In this paper, we propose a novel framework, titled AutoCNN based on artificial intelligence techniques, to predict future stock market indexes. AutoCNN is composed mainly of three stages: (1) CNN for Automatic Feature Extraction, (2) The Halving Grid Search algorithm is combined to CNN model for stock indexes prediction and (3) Evaluation and recommendation. To validate our AutoCNN, we conduct experiments on two financial datasets that are extracted in the period between 2018 and 2023, which includes several events such as economic, health and geopolitical international crises. The performance of AutoCNN model is quantified using various metrics. It is benchmarked against different models and it proves strong prediction abilities. AutoCNN contributes to emerging technologies and innovation in the financial sector by automating decision-making, leveraging advanced pattern recognition, and enhancing the overall decision support system for investors in the digital economy.

Résumé

The stock market which is a particular type of financial market, has attracted nowadays the attention of financial analysts and investors, as it is recognized one of the most challenging and unpredictable market. Recently, this kind of market is well known by its extreme volatility and instability due to the health (COVID-19), the geopolitical (the Russian, Ukraine, European, and American conflict), and the economic crises. This situation intensified the uncertainty and fear of investors; they need an intelligent and stable decision support system to assist them to foresee the future variations of stock prices. To address this issue, our paper proposes a hybrid ensemble-learning model that integrates different methods. (1) The Singular Spectrum Analysis (SSA) is used to eliminate the noise from financial data. (2) The Convolutional Neural Network (CNN) is applied to handle the task of automatic feature extraction. (3) Three machine-learning predictors (Random Forest (RF), Gradient Boosting (GB), and Extra Trees (ET)) are merged together and optimized using the Halving Grid Search (HGS) algorithm to obtain collective final predictions. To verify the validity of the proposed model, two major indices of Chinese and American stock markets, namely SSE and NASDAQ, were used. The proposed model is evaluated using RMSE, MAE, MAPE and CPU time metrics. Based on experiments, it is proven that the achieved results are better than other comparative prediction models used for benchmarking.

Résumé

Financial markets operate as dynamic systems susceptible to ongoing changes influenced by recent crises, such as geopolitical and health crises. Due to these factors, investor uncertainty has increased, making it challenging to identify trends in the stock markets. Predicting stock market prices enhances investors’ ability to make accurate investment decisions. This paper proposes an intelligent financial system named Pred-IFDSS, aiming to recommend the best model for accurate predictions of future stock market indexes. Pred-IFDSS includes seven machine learning models: (1) Linear Regression (LR), (2) Support Vector Regression (SVR), (3) eXtreme Gradient Boosting (XGBoost), (4) Simple Recurrent Neural Network (SRNN), (5) Gated Recurrent Unit (GRU), (6) Long Short-Term Memory (LSTM), and (7) Artificial Neural Network (ANN). Each model is tuned using the grid search strategy, trained, and evaluated. Experiments are conducted on three stock market indexes (NASDAQ, S&P 500, and NYSE). To measure the performance of these models, three standard strategic indicators are employed (MSE, RMSE, and MAE). The outcomes of the experiments demonstrate that the error rate in SRNN model is very low, and we recommend it to assist investors in foreseeing future trends in stock market prices and making the right investment decisions.

Résumé

This paper introduces TradeStrat-ML, a novel framework for stock market trading. It integrates various techniques: technical analysis, hybrid machine learning models, multi-agent-based simulations (MABS), and financial modeling for stock market analysis and future predictions. The process involves using a Convolutional Neural Network (CNN) to extract features from preprocessed financial data. The output of this model is then combined with three machine learning models (Gated Recurrent Unit (GRU), Extreme Gradient Boosting (XGBoost), and Support Vector Regression (SVR)) to predict future stock price indices. Subsequently, the models are evaluated, compared, and the most accurate model is selected for stock market prediction. In the final stage, the selected model, along with the Simple Moving Average (SMA) indicator, is used to develop an optimized trading strategy. The TradeStrat-ML system is organized into four main layers and validated using MABS simulations. Comparative analysis and simulation experiments collectively indicate that this new combination prediction model is a potent and practical tool for informed investment decision-making.

Résumé

Knowledge discovery from mobility data is about identifying behaviours from trajectories. In fact, mining masses of trajectories is required to have an overview of this data, notably, investigate the relationship between different entities movement. Most state-of-the-art work in this issue operates on raw trajectories. Nevertheless, behaviours discovered from raw trajectories are not as rich and meaningful as those discovered from semantic trajectories. In this paper, we establish a mining approach to extract patterns from semantic trajectories. We propose to apply sequential pattern mining based on a pre-processing step of clustering to alleviate the former's temporal complexity. Mining considers the spatial and temporal dimensions at different levels of granularity providing then richer and more insightful patterns about humans behaviour. We evaluate our work on tourists semantic trajectories in Kyoto. Results showed the effectiveness and efficiency of our model compared to state-of-the-art work.

Résumé

The continuous development of Internet infrastructures and the evolution of digital electronics, particularly Nano-computers, are making the Internet of Things (IoT) emergent. Despite the progress, these IoT objects suffer from a crucial problem which is their limited power supply. IoT objects are often deployed as an ad-hoc network. To minimize their consumption of electrical energy, clustering techniques are used. In this paper, a centralized clustering algorithm with single-hop routing based on a genetic algorithm and Improved k-means is proposed. The proposed approach is compared with the LEACH, K-means and OK-means algorithms. Simulation results show that the proposed algorithm performs well in terms of network lifetime and energy consumption.

Résumé

Supervised machine learning techniques include classification and regression. In regression, the objective is to map a real-valued output to a set of input features. The main challenge that existing methods for regression encounter is how to maintain an accuracy-simplicity balance. Since Regression Trees (RTs) are simple to interpret, many existing works have focused on proposing RT and Model Tree (MT) induction algorithms. MTs are RTs with a linear function at the leaf nodes rather than a numerical value are able to describe the relationship between the inputs and the output. Traditional RT induction algorithms are based on a top-down strategy which often leads to a local optimal solution. Other global approaches based on Evolutionary Algorithms (EAs) have been proposed to induce RTs but they can require an important calculation time which may affect the convergence of the algorithm to the solution. In this paper, we introduce a novel approach called Bi-level Evolutionary Model Tree Induction algorithm for regression, that we call BEMTI, and which is able to induce an MT in a bi-level design using an EA. The upper-level evolves a set of MTs using genetic operators while the lower-level optimizes the Linear Models (LMs) at the leaf nodes of each MT in order to fairly and precisely compute their fitness and obtain the optimal MT. The experimental study confirms the outperformance of our BEMTI compared to six existing tree induction algorithms on nineteen datasets.

Résumé

Elderly fatigue, a critical issue affecting the health and well-being of the aging population worldwide, presents as a substantial decline in physical and mental activity levels. This widespread condition reduces the quality of life and introduces significant hazards, such as increased accidents and cognitive deterioration. Therefore, this study proposed a model to detect fatigue in the elderly with satisfactory accuracy. In our contribution, we use video and image processing through a video in order to detect the elderly’s face recognition in each frame. The model identifies facial landmarks on the detected face and calculates the Eye Aspect Ratio (EAR), Eye Fixation, Eye Gaze Direction, Mouth Aspect Ratio (MAR), and 3D head pose. Among the various methods evaluated in our study, the Extra Trees algorithm outperformed all others machine learning methods, achieving the highest results with a sensitivity of 98.24%, specificity of 98.35%, and an accuracy of 98.29%.

Résumé

Sleep is a vital process for human well-being. Sleep scoring is performed by experts using polysomnograms, that record several body activities, such as electroencephalograms (EEG), electrooculograms (EOG), and electromyograms (EMG). This task is known to be exhausting, biased, time-consuming, and prone to errors. Current automatic sleep scoring approaches are mostly based on single-channel EEG and do not produce explainable results. Therefore, we propose a heterogeneous ensemble learning-based approach where we combine accuracy-based learning classifier systems with different algorithms to produce a robust, explainable, and enhanced classifier. The efficiency of our approach was evaluated using the Sleep-EDF benchmark dataset. The proposed models have reached an accuracy of 89.2% for the stacking model and 87.9% for the voting model, on a multi-class classification task based on the R&K guidelines.

Résumé

Financial markets are dynamic and open systems. They are subject to the influence of environmental changes. For this reason, predicting stock market prices is a difficult task for investors due to the volatility of the financial stock markets nature. Stock market forecasting leads investors to make decisions with more confidence based on the prediction of stock market price behavior. Indeed, a lot of analysts are greatly focused in the research domain of stock market prediction. Generally, the stock market prediction tools are categorized into two types of algorithms: (1) linear models like Auto Regressive (AR), Moving Average (MA), Auto-Regressive Integrated Moving Average (ARIMA), and (2) non-linear models like Autoregressive Conditionally Heteroscedastic (ARCH), Generalized Autoregressive Conditional Heteroskedasticity (GARCH) and recently Neural Network (NN)). This paper aspires to crucially predict the stock index movement for National Association of Securities Dealers Automated Quotations (NASDAQ) based on deep learning networks. We propose a hybrid stock price prediction model using Convolutional Neural Network (CNN) for feature selection and Neural Network models to perform the task of prediction. To evaluate the performance of the proposed models, we use five regression evaluation metrics: Root Mean Square Error (RMSE), Mean Square Error (MSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE) and R-Square (R2), and the Execution Time (ET) metric to calculate the necessary time for running each hybrid model. The results reveal that error rates in the CNN-BGRU model are found to be lower compared to CNN-GRU, CNN-LSTM, CNN-BLSTM and the the existing hybrid models. This research work produces a practical experience for decision makers on financial time series data.

Résumé

The prediction of stock price movements is one of the most challenging tasks in financial market field. Stock price trends depended on various external factors like investor's sentiments, health and political crises which can make stock prices more volatile and chaotic. Lately, two crises affected the variation of stock prices, COVID-19 pandemic and Russia-Ukraine conflict. Investors need a robust system to predict future stock trends in order to make successful investments and to face huge losses in uncertainty situations. Recently, various machine learning (ML) models have been proposed to make accurate stock movement predictions. In this paper, a framework including five ML classifiers (Gaussian Naive Bayes (GNB), Random Forest (RF), Gradient Boosting (GB), Support Vector Machine (SVM), and K-Nearest Neighbors (kNN))) is proposed to predict the closing price trends. Technical indicators are calculated and used with historical stock data as input. These classifiers are hybridized with Principal Component Analysis method (PCA) for feature selection and Grid Search (GS) Optimization Algorithm for hyper-parameters tuning. Experimental results are conducted on National Association of Securities Dealers Automated Quotations (NASDAQ) stock data covering the period from 2018 to 2023. The best result was found with the Random Forest classifier model which achieving the highest accuracy (61%).

Résumé

Fake news has a worldwide impact and the potential to change political scenarios and human behavior, especially in a critical time like the COVID-19 pandemic. This work suggests a Soft Three-Level Voting Model (Soft T-LVM) for automatically classifying COVID-19 fake news. We train different individual machine learning algorithms and different ensemble methods in order to overcome the weakness of individual models. This novel model is based on the soft-voting technique to calculate the class with the majority of votes and to choose the classifiers to merge and apply at every level. We use the Grid search method to tune the hyper-parameters during the process of classification and voting. The experimental evaluation confirms that our proposed model approach has superior performance compared to the other classifiers.

Résumé

Identifying opinion leaders in social networks, particularly in social media, is a crucial marketing strategy. These individuals have a considerable influence on the purchasing decisions of their communities. Companies can benefit from collaborating with relevant opinion leaders in their market as this can increase their visibility, establish their credibility, and gain consumer trust, leading to increased sales, improved brand perception, and an expanded market share. Additionally, by gaining a comprehensive understanding of opinion leaders, companies can better comprehend the trends and preferences of their target audience. This allows them to tailor their marketing and product strategies more effectively. Identifying suitable influencers to endorse their products or services is a significant challenge for companies. The identification of opinion leaders is complicated by their informal and unstructured nature, as well as the varying selection criteria depending on the marketing campaign’s goals. While numerous research studies have focused on detecting opinion leaders in social networks based on content, interactions, or a combination of both, few have explored sentiment analysis of post content, received interactions, and user comments in relation to published posts. The purpose of this paper is to present an hybrid approach to detect opinion leaders in Facebook. This approach involves analyzing the trajectory of post content by examining interactions on the post, as well as mining the text content of the post itself and analyzing the users’comments sentiments.

Résumé

Résumé

This paper presents a new evolutionary method and tool called BMLDS (Bi-level Multi-Label Detection of Smells) that optimizes a population of classifier chains for the multi-label detection of smells. As the chain is sensitive to the labels' (i.e., smell types) order, the chains induction task is framed as a bi-level optimization problem, where the upper-level role is to search for the optimal order of each considered chain while the lower-level one is to generate the chains. This allows taking into consideration the interactions between smells in the multi-label detection process. The statistical analysis of the experimental results reveals the merits of our proposal with respect to several existing works.

Résumé

Code smells, also known as anti-patterns, are poor design or implementation choices that hinder program comprehensibility and maintainability. While several code smell detection methods have been proposed, Mantyla et al. identified the uncertainty issue as one of the major individual human factors that may affect developer’s decisions about the smelliness of software classes: they may indeed have different opinions mainly due to their different knowledge and expertise. Unfortunately, almost all the existing approaches assume data perfection and neglect the uncertainty when identifying the labels of the software classes. Ignoring or rejecting any uncertainty form could lead to a considerable loss of information, which could significantly deteriorate the effectiveness of the detection and identification processes. Inspired by our previous works and motivated by the interesting performance of the PDT (Possibilistic Decision Tree) in classifying uncertain data, we propose ADIPE (Anti-pattern Detection and Identification using Possibilistic decision tree Evolution), as a new tool that evolves and optimizes a set of detectors (PDTs) that could effectively deal with software class labels uncertainty using some concepts from the Possibility theory. ADIPE uses a PBE (Possibilistic Base of Examples: a dataset with possibilistic labels) that it is built using a set of opinion-based classifiers (i.e., a set of probabilistic classifiers) with the aim to simulate human developers’ uncertainty. A set of advisors and probabilistic classifiers are employed in order to mimic the subjectivity and the doubtfulness of software engineers. A detailed experimental study is conducted to show the merits and outperformance of ADIPE in dealing with uncertainty in code smells detection and identification with respect to four relevant state-of-the-art methods, including the baseline PDT. The experimental study was performed in uncertain and certain environments based on two suitable metrics: PF-measure_dist (Possibilistic F-measure_Distance) and IAC (Information Affinity Criterion); which corresponds to the F-measure and Accuracy (PCC) for the certain case. The obtained results for the uncertain environment reveal that for the detection process, the PF-measure_dist of ADIPE ranges within [0.9047 and 0.9285], and its IAC lies within [0.9288 and 0.9557]; while for the identification process, the PF-measure_dist of ADIPE is in [0.8545, 0.9228], and its IAC lies within [0.8751, 0.933]. ADIPE is able to find 35% more code smells with uncertain data than the second best algorithm (i.e., BLOP). In addition, ADIPE succeeds to decrease the number of false alarms (i.e., misclassified smelly instances) with a rate equals to 12%. Our proposed approach is also able to identify 43% more smell types than BLOP and decreases the number of false alarms with a rate equals to 32%. Similar results were obtained for the certain environment, which demonstrate the ability of ADIPE to also deal with the certain environment.

Résumé

Résumé

In this study, various organizations that have participated in several road path-detecting experimentsare analyzed. However, the majority of techniques rely on attributes or form models built by humans to identifysections of the path. In this paper, a suggestion was made regarding a road path recognition structure that is dependenton a deep convolutional neural network. A tiny neural network has been developed to perform feature extraction toa massive collection of photographs to extract the suitable path feature. The parameters obtained from the model ofthe route classification network are utilized in the process of establishing the parameters of the layers that constitutethe path detection network. The deep convolutional path discovery network’s production is pixel-based and focuseson the identification of path types and positions. To train it, a detection failure job is provided. Failure in pathclassification and regression are the two components that make up a planned detection failure function. Instead oflaborious postprocessing, a straightforward solution to the problem of route marking can be found using observedpath pixels in conjunction with a consensus of random examples. According to the findings of the experiments, theclassification precision of the network for classifying every kind is higher than 98.3%. The simulation that was trainedusing the suggested detection failure function is capable of achieving an accuracy of detection that is 85.5% over atotal of 30 distinct scenarios on the road

Résumé

Cost-sensitive learning is one of the most adopted approaches to deal with data imbalance in classification. Unfortunately, the manual definition of misclassification costs is still a very complicated task, especially with the lack of domain knowledge. To deal with the issue of costs' uncertainty, some researchers proposed the use of intervals instead of scalar values. This way, each cost would be delimited by two bounds. Nevertheless, the definition of these bounds remains as a very complicated and challenging task. Recently, some researches proposed the use of genetic programming to simultaneously build classification trees and search for optimal costs' bounds. As for any classification tree there is a whole search space of costs' bounds, we propose in this paper a bi-level evolutionary approach for interval-based cost-sensitive classification tree induction where the trees are constructed at the upper level while misclassification costs intervals bounds are optimized at the lower level. This ensures not only a precise evaluation of each tree but also an effective approximation of optimal costs intervals bounds. The performance and merits of our proposal are shown through a detailed comparative experimental study on commonly used imbalanced benchmark data sets with respect to several existing works.

Résumé

In recent years, Home Health Care (HHC) has gained popularity in different countries around the world (e.g. France, US, Germany, etc.). The HHC consists in providing medical services to patients at home. During the HHC service, caregivers’ planning may be disrupted by some unexpected events (e.g. urgent request, caregiver absence, traffic congestion, etc.), which makes HHC activities infeasible. This paper addresses the daily HHC routing and scheduling problem by considering unpredicted events. To solve this problem, we propose a Case-Based Reasoning (CBR) methodology. Our purpose is to create the HHC case base which contains the knowledge about the perturbation.

Résumé

Recent researches on computational modeling show that emotions have a major influence on human behavior and decision making. Therefore, it is recognized that they are necessary to produce human-like in artificial agents. Several computational behavior models have been proposed. However, some of them have incorporated the emotion, others have integrated the psychological aspects in order to study the human behavior, but they did not take into account both of the emotional and the psychological impacts. In this context, we attempt to present an overview of the existent works. Then, we aim to present a new behavior agent model that integrates both of the psychological and emotional aspects to prove their impacts on the human decision.

Résumé

The aim of this paper is to classify the driver tiredness and fatigue in automobile via eye tracking and body movements using deep learning based Convolutional Neural Network (CNN) algorithm. Vehicle driver face localization serves as one of the most widely used real-world applications in fields like toll control, traffic accident scene analysis, and suspected vehicle tracking. The research proposed a CNN classifier for simultaneously localizing the region of human face and eye positioning. The classifier, rather than bounding rectangles, gives bounding quadrilaterals, which gives a more precise indication for vehicle driver face localization. The adjusted regions are preprocessed to remove noise and passed to the CNN classifier for real time processing. The preprocessing of the face features extracts connected components, filters them by size, and groups them into face expressions. The employed CNN is the well-known technology for human face recognition. One we aim to extract the facial landmarks from the frames, we will then leverage classification models and deep learning based convolutional neural networks that predict the state of the driver as 'Alert' or 'Drowsy' for each of the frames extracted. The CNN model could predict the output state labels (Alert/Drowsy) for each frame, but we wanted to take care of sequential image frames as that is extremely important while predicting the state of an individual. The process completes, if all regions have a sufficiently high score or a fixed number of retries are exhausted. The output consists of the detected human face type, the list of regions including the extracted mouth and eyes with recognition reliability through CNN with an accuracy of 98.57% with 100 epochs of training and testing.

Résumé

Smart health-care assistants are designed to improve the comfort of the patient where smart refers to the ability to imitate the human intelligence to facilitate his life without, or with limited, human intervention. As a part of this, we are proposing a new Intelligent Communication Assistant capable of detecting physiological needs by following a new efficient Inverse Reinforcement learning algorithm designed to be able to deal with new time-recorded states. The latter processes the patient’s environment data, learns from the patient previous choices and becomes capable of suggesting the right action at the right time. In this paper, we took the case study of Locked-in Syndrome patients, studied their actual communication methods and tried to enhance the existing solutions by adding an intelligent layer. We showed that by using Deep Inverse Reinforcement Learning using Maximum Entropy, we can learn how to regress the reward amount of new states from the ambient environment recorded states. After that, we can suggest the highly rewarded need to the target patient. Also, we proposed a full architecture of the system by describing the pipeline of the information from the ambient environment to the different actors.

Résumé

Code smells (a.k.a. anti-patterns) are manifestations of poor design solutions that could deteriorate the software maintainability and evolution. Despite the high number of existing detection methods, the issue of class label uncertainty is usually omitted. Indeed, two human experts may have different degrees of uncertainty about the smelliness of a particular software class not only for the smell detection task but also for the smell type identification one. Thus, this uncertainty should be taken into account and then processed by detection tools. Unfortunately, these latter usually reject and/or ignore uncertain data that correspond to software classes (i.e. dataset instances) with uncertain labels. This practice could considerably degrade the detection/identification process effectiveness. Motivated by this observation and the interesting performance of the Possibilistic K-NN (PK-NN) classifier in dealing with uncertain data, we propose a new possibilistic evolutionary detection method, named ADIPOK (Anti-patterns Detection and Identification using Possibilistic Optimized K-NNs), that is able to deal with label uncertainty using some concepts stemming from the Possibility theory. ADIPOK is validated using a possibilistic base of smell examples that simulates the subjectivity of software engineers’ opinions’ uncertainty. The statistical analysis of the obtained results on a set of comparative experiments with respect to four state-of-the-art methods show the merits of our proposed method.

Résumé

Like the case of any software, Web Services (WSs) developers could introduce anti-patterns due to the lack of experience and badly-planned changes. During the last decade, search-based approaches have shown their outperformance over other approaches mainly thanks to their global search ability. Unfortunately, these approaches do not consider the uncertainty of class labels. In fact, two experts could be uncertain about the smelliness of a particular WS interface but also about the smell type. Currently, existing works reject uncertain data that correspond to WSs interfaces with doubtful labels. Motivated by this observation and the good performance of the possibilistic K-NN classifier in handling uncertain data, we propose a new evolutionary detection approach, named Web Services Anti-patterns Detection and Identification using Possibilistic Optimized K-NNs (WS-ADIPOK), which can cope with the uncertainty based on the Possibility Theory. The obtained experimental results reveal the merits of our proposal regarding four relevant state-of-the-art approaches.

Résumé

Résumé

Résumé

Code smells, also known as anti-patterns, are indicators of bad design solutions. However, two different experts may have different opinions not only about the smelliness of a particular software class but also about the smell type. This causes an uncertainty problem that should be taken into account. Unfortunately, existing works reject uncertain data that correspond to software classes with doubtful labels. Uncertain data rejection could cause a significant loss of information that could considerably degrade the performance of the detection process. Motivated by this observation and the good performance of the possibilistic K-NN classifier in handling uncertain data, we propose in this paper a new evolutionary detection method, named ADIPOK (Anti-pattern Detection and Identification using Possibilistic Optimized K-NN), that is able to cope with the uncertainty factor using the possibility theory. The comparative experimental results reveal the merits of our proposal with respect to four relevant state-of-the-art approaches.

Résumé

Recent advances in Computational Intelligent Systems have focused on addressing complex problems related to the dynamicity of the environments. Generally in dynamic environments, data are presented as streams that may evolve over time and this is known by concept drift. Handling concept drift through ensemble classifiers has received a great interest in last decades. The success of these ensemble methods relies on their diversity. Accordingly, various diversity techniques can be used like block-based data, weighting-data or filtering-data. Each of these diversity techniques is efficient to handle certain characteristics of drift. However, when the drift is complex, they fail to efficiently handle it. Complex drifts may present a mixture of several characteristics (speed, severity, influence zones in the feature space, etc.) which may vary over time. In this case, drift handling is more complicated and requires new detection and updating tools. For this purpose, a new ensemble approach, namely EnsembleEDIST2, is presented. It combines the three diversity techniques in order to take benefit from their advantages and outperform their limits. Additionally, it makes use of EDIST2, as drift detection mechanism, in order to monitor the ensemble’s performance and detect changes. EnsembleEDIST2 was tested through different scenarios of complex drift generated from synthetic and real datasets. This diversity combination allows EnsembleEDIST2 to outperform similar ensemble approaches in terms of accuracy rate, and present stable behaviors in handling different scenarios of complex drift.

Résumé

Bi-Level Optimization Problem (BLOP) is a class of challenging problems with two levels of optimization tasks. The main goal is to optimize the upper level problem, which has another optimization problem as a constraint. In this way, the evaluation of each upper level solution requires finding an optimal solution to the corresponding lower level problem, which is computationally so expensive. For this reason, most proposed bi-level resolution methods have been restricted to solve the simplest case (linear continuous BLOPs). This fact has attracted the evolutionary computation community to solve such complex problems. Besides, to enhance the search performance of Evolutionary Algorithms (EAs), reusing knowledge captured from past optimization experiences along the search process has been proposed in the literature, and was demonstrated much promise. Motivated by this observation, we propose in this paper, a memetic version of our proposed Co-evolutionary Decomposition-based Algorithm-II (CODBA-II), that we named M-CODBA-II, to solve combinatorial BLOPs. The main motivation of this paper is to incorporate transfer learning within our recently proposed CODBA-II scheme to make the search process more effective and more efficient. Our proposed hybrid algorithm is investigated on two bi-level production-distribution problems in supply chain management formulated to: (1) Bi-CVRP and (2) Bi-MDVRP. The experimental results reveal a potential advantage of memes incorporation in CODBA-II. Most notably, the results emphasize that transfer learning allows not only accelerating the convergence but also finding better solutions.

Résumé

During the last decades, Learning Classifier Systems have known many advancements that were highlighting their potential to resolve complex problems. Despite the advantages offered by these algorithms, it is important to tackle other aspects such as the uncertainty to improve their performance. In this paper, we present a new Learning Classifier System (LCS) that deals with uncertainty in the class selection in particular imprecision. Our idea is to integrate the Belief function theory in the sUpervised Classifier System (UCS) for classification purpose. The new approach proved to be efficient to resolve several classification problems.

Résumé

During the last decades, Learning Classifier Systems have known many advancements that were highlighting their potential to resolve complex problems. Despite the advantages offered by these algorithms, it is important to tackle other aspects such as the uncertainty to improve their performance. In this paper, we present a new Learning Classifier System (LCS) that deals with uncertainty in the class selection in particular imprecision. Our idea is to integrate the Belief function theory in the sUpervised Classifier System (UCS) for classification purpose. The new approach proved to be efficient to resolve several classification problems.

Résumé

Recent advances in computational intelligent systems have focused on addressing complex problems related to the dynamicity of the environments. In increasing number of real world applications, data are presented as streams that may evolve over time and this is known by concept drift. Handling concept drift is becoming an attractive topic of research that concerns multidisciplinary domains such that machine learning, data mining, ubiquitous knowledge discovery, statistic decision theory, etc... Therefore, a rich body of the literature has been devoted to the study of methods and techniques for handling drifting data. However, this literature is fairly dispersed and it does not define guidelines for choosing an appropriate approach for a given application. Hence, the main objective of this survey is to present an ease understanding of the concept drift issues and related works, in order to help researchers from different disciplines to consider concept drift handling in their applications. This survey covers different facets of existing approaches, evokes discussion and helps readers to underline the sharp criteria that allow them to properly design their own approach. For this purpose, a new categorization of the existing state-of-the-art is presented with criticisms, future tendencies and not-yet-addressed challenges.

Résumé

Detecting changes in data streams attracts major attention in cognitive computing systems. The challenging issue is how to monitor and detect these changes in order to preserve the model performance during complex drifts. By complex drift, we mean a drift that presents many characteristics in the sometime. The most challenging complex drifts are gradual continuous drifts, where changes are only noticed during a long time period. Moreover, these gradual drifts may also be local, in the sense that they may affect a little amount of data, and thus make the drift detection more complicated. For this purpose, a new drift detection mechanism, EDIST2, is proposed in order to deal with these complex drifts. EDIST2 monitors the learner performance through a self-adaptive window that is autonomously adjusted through a statistical hypothesis test. This statistical test provides theoretical guarantees, regarding the false alarm rate, which were experimentally confirmed. EDIST2 has been tested through six synthetic datasets presenting different kinds of complex drift, and five real-world datasets. Encouraging results were found, comparing to similar approaches, where EDIST2 has achieved good accuracy rate in synthetic and real-world datasets and has achieved minimum delay of detection and false alarm rate.

Résumé

Detecting changes in data streams attracts major attention in cognitive computing systems. The challenging issue is how to monitor and detect these changes in order to preserve the model performance during complex drifts. By complex drift, we mean a drift that presents many characteristics in the sometime. The most challenging complex drifts are gradual continuous drifts, where changes are only noticed during a long time period. Moreover, these gradual drifts may also be local, in the sense that they may affect a little amount of data, and thus make the drift detection more complicated. For this purpose, a new drift detection mechanism, EDIST2, is proposed in order to deal with these complex drifts. EDIST2 monitors the learner performance through a self-adaptive window that is autonomously adjusted through a statistical hypothesis test. This statistical test provides theoretical guarantees, regarding the false alarm rate, which were experimentally confirmed. EDIST2 has been tested through six synthetic datasets presenting different kinds of complex drift, and five real-world datasets. Encouraging results were found, comparing to similar approaches, where EDIST2 has achieved good accuracy rate in synthetic and real-world datasets and has achieved minimum delay of detection and false alarm rate.

Résumé

Detecting and monitoring changes during the learning process are important areas of research in many industrial applications. The challenging issue is how to diagnose and analyze these changes so that the accuracy of the learning model can be preserved. Recently, ensemble classifiers have achieved good results when dealing with concept drifts. This paper presents two ensembles learning algorithms BagEDIST and BoostEDIST, which respectively combine the Online Bagging and the Online Boosting with the drift detection method EDIST. EDIST is a new drift detection method which monitors the distance between two consecutive errors of classification. The idea behind this combination is to develop an ensemble learning algorithm which explicitly handles concept drifts by providing useful descriptions about location, speed and severity of drifts. Moreover, this paper presents a new drift diversity measure in order to study the diversity of base classifiers and see how they cope with concept drifts. From various experiments, this new measure has provided a clearer vision about the ensemble’s behavior when dealing with concept drifts.

Résumé

La classification dynamique s’intéresse au traitement des données non-stationnaires issues des environnements évolutifs dans le temps. Ces données peuvent présenter des dérives, qui affectent la performance du modèle d’apprentissage initialement construit. Aujourd’hui, beaucoup d’intérêts sont portés sur la surveillance, la mise à jour et le diagnostic de ces dérives afin d’améliorer la performance du modèle d’apprentissage. Dans ce contexte, une nouvelle méthode de détection de dérive basée sur la distance entre les erreurs de classification est présentée. Cette méthode, nommée EDIST, surveille la distribution des distances des erreurs de classification entre deux fenêtres de données afin de détecter une différence à travers un test d’hypothèse statistique. EDIST a été testée à travers des bases de données artificielles et réelles. Des résultats encourageants ont été trouvés par rapport à des méthodes similaires. EDIST a pu trouver les meilleurs taux d’erreur de classification dans la plupart des cas et a montré une robustesse envers le bruit et les fausses alarmes.

Résumé

Learning agents have to deal with the exploration-exploitation
dilemma. The choice between exploration and exploitation is very difficult in
dynamic systems; in particular in large scale ones such as economic systems.
Recent research shows that there is neither an optimal nor a unique solution for
this problem. In this paper, we propose an adaptive approach based on meta-rules
to adapt the choice between exploration and exploitation. This new adaptive approach relies on the variations of the performance of the agents. To validate the
approach, we apply it to economic systems and compare it to two adaptive methods: one local and one global. Herein, we adapt these two methods, which were
originally proposed by Wilson, to economic systems. Moreover, we compare different exploration strategies and focus on their influence on the performance of
the agents

Résumé

Economic markets are complex systems. They are characterized by a large and dynamic population of firms. To deal with this complexity, we propose an adaptive multiagent system which models a set of firms in competition with each other within a shared market. The firms are represented by agents; each firm is represented by an adaptive agent. We show the advantages of adaptive agents to represent firms. Moreover, we underline the limits of the economic models which account for the firms only and ignore the organizational forms. We propose a new adaptive multiagent model that includes the organizational forms into the economic models. We simulate this model and discuss its advantages.