Résumé: Machine learning algorithms have offered unprecedented solutions for many real-world problems. These algorithms frequently involve using a large number of features. However, several of these features could not be very informative due to data uncertainties, such as noise and residual variation. Decision trees are among the most preferred classification models. This is due to their simplicity, explainability, and readability. However, data inaccuracies could impact the construction of decision trees and thus hinder their results. Feature selection and construction present promising research direction to enhance the performance of decision tree models. In this paper, we present a strategy that combines feature selection and construction where the construction of new features is performed by using the ones that were not chosen during the selection step. However, the search space of combinations of selected/constructed features is extremely large. To find the best solution, a genetic algorithm has been developed combined with a graph covering vertices set guided approach. The obtained results on a large number of datasets from the UCI Repository demonstrate that our approach outperforms both recent and classical decision tree construction techniques. We also present a successful use case of our approach in detecting Botnet traffic in the Internet of Vehicles.

Résumé: In this study, we introduce a novel local fractional integral identity related to the Gaussian two-point left Radau rule. Based on this identity, we establish some new fractal inequalities for functions whose first-order local fractional derivatives are generalized convex and concave. The obtained results not only represent an extension of certain previously established findings to fractal sets but also a refinement of these when the fractal dimension ? is equal to one. Finally, to support our findings, we present a practical application to demonstrate the effectiveness of our results.

Résumé: Random Forest is an ensemble classification approach. It aims to design a discrete finite group of decision trees constructed based on bootstrap samples and random attribute selection. Random Forests have strong generalization capacities due to the variance in the training and attribute couple subsets used for constructing different decision trees in the forest. However, to construct a robust and effective random forest, two main issues need to be taken into account namely: (1) increasing the accuracy and diversity of decision trees; (2) decreasing the number of decision trees. In this paper, a genetic algorithm-based approach to tackle the aforementioned challenges related to random forest construction is proposed. Three objectives are taken into consideration. First, strengthening the classification accuracy of individual decision trees as well as that of the forest. Second, making use of diversity measures among the decision trees to improve the generalization of the constructed model. Third, minimizing the number of trees in the forest and finding an optimal subset of the random forest. An experimental evaluation on several datasets from the UCI Machine Learning Repository is conducted. The obtained results show that the proposed approach outperforms state-of-the-art classical as well as evolutionary random forest construction methods. Finally, the proposed approach is used to build a reliable random forest model for detecting Botnet traffic in Internet of Things environment.

Résumé: For memory-based collaborative filtering, the quality of the target user's neighbourhood plays an important role for providing him/her with successful item recommendations. The existent techniques for neighbourhood selection aim to maximise the pairwise similarity between the target user and his/her neighbours, which mainly improves only the recommendation accuracy. However, these methods do not consider other important aspects for successful recommendations such as providing diversified and novel item recommendations, which also highly affect users' satisfaction. In this paper, we linearly combine two probabilistic criteria for selecting the right neighbourhood of a target user and provide him/her accurate, and novel item recommendations. The combination of these two probabilistic quality measures forms a fitness function, which guides the evolution of a genetic algorithm. For each target user, the genetic algorithm explores the user's whole search space and selects the most suitable neighbourhood. Thus, our approach makes a balance between the accuracy and the novelty of the provided item recommendations, as will be experimentally shown on MovieLens dataset.

Résumé: Recommender systems aim to provide users with recommendations of quality. New evaluation metrics such as diversity, have taken an increasing interest in a wide spectrum of applications, including the ecommerce, due to their ability to improve online revenues. High recommendation diversity allows a higher chance to satisfy the users’ needs. However, in a large market of users and products, the scalability of the system is questionable because of the required computing resources. We present a scalable evolutionary clustering algorithm that allows to target two objectives. The proposed solution balances between the recommendation accuracy and coverage by making an overlapped clustering. In our approach, we use a Genetic Algorithm to assign each user to a main cluster from which he gets his recommendations and to secondary clusters as a candidate neighbor. The performance comparison of our algorithm against classic well-known approaches, such as k-NN based Collaborative Filtering, showed a significant improvement.

Résumé: The Advanced Metering Infrastructure (AMI) is one of the key components of the smart grid. It provides interactive services for managing billing and electricity consumption, but it also introduces new vectors for cyberattacks. Although, the devastating and severe impact of power overloading cyberattacks on smart grid AMI, few researches in the literature have addressed them. In the present paper, we propose a two-level anomaly detection framework based on regression decision trees. The introduced detection approach leverages the regularity and predictability of energy consumption to build reference consumption patterns for the whole neighborhood and each household within it. Using a reference consumption pattern enables detecting power overloading cyberattacks regardless of the attacker’s strategy as they cause a drastic change in the consumption pattern. The continuous two-level monitoring of energy consumption load allows efficient and early detection of cyberattacks. We carried out an extensive experiment on a real-world publicly available energy consumption dataset of 500 customers in Ireland. We extracted, from the raw data, the relevant attributes for training the energy consumption patterns. The evaluation shows that our approach achieves a high detection rate, a low false alarm rate, and superior performances compared to existing solutions.

Résumé: Graph clustering is one of the most important research topics in graph mining and network analysis. Given the abundance of data in many real-world applications, graph nodes and edges could be annotated with multiple sets of attributes that could be derived from heterogeneous data sources. The consideration of these attributes during graph clustering would facilitate the generation of graph clusters with balanced and cohesive intra-cluster structures and nodes with homogeneous properties. In this paper, we propose a graph clustering approach for mining skyline clusters over large attributed graphs based on the dominance relationship. Each skyline solution is optimized simultaneously for multiple fitness functions, each function is defined over the graph topology or over a particular set of attributes derived from multiple data sources. We evaluate our approach experimentally with a large protein-protein interaction network of the human interactome enriched with large sets of heterogeneous cancer-associated attributes. The results demonstrate the efficiency of our approach and show how integrating node attributes from multiple data sources can result in a more robust graph clustering than the consideration of the graph topology alone.

Résumé: Decision trees are efficient means for building classification models due to the compressibility, simplicity and ease of interpretation of their results. However, during the construction phase of decision trees, the outputs are often large trees that are affected by many uncertainties in the data (particularity, noise and residual variation). Combining attribute selection and data sampling presents one of the most promising research directions to overcome decision tree construction problems. However, the search space composed of all possible combinations of subsets of training samples and attributes is extremely large. In this paper, a novel approach is presented that allows generating an optimized decision tree by selecting an optimal couple of training samples and attributes subsets for training. As the search space of candidate couples of training samples and attributes subsets is extremely large, we use particle swarm optimization to make the search of an “optimal” solution tractable. The selected optimized solution helps in avoiding over-fitting and complexity problems suffered in the construction phase of decision trees. We conducted an extensive experimental evaluation on 22 datasets from the UCI Machine Learning Repository. The obtained results show that the proposed approach outperforms state-of-the-art classical as well as evolutionary decision tree construction methods in terms of simplicity, accuracy, and F-measure. We further evaluate our approach on a real-world engineering application for condition monitoring of rotating machinery under severe non-stationary conditions. The obtained results showed that the proposed approach allowed to optimize the use of instantaneous angular speed to diagnose gears defects.

Résumé: The primary task of a memory-based collaborative filtering (CF) recommendation system is to select a group of nearest (similar) user neighbors for an active user. Traditional memory-based CF schemes tend to only focus on improving as much as possible the accuracy by recommending familiar items (i.e., popular items over the group). Yet, this may reduce the number of items that could be recommended and consequently weakens the chances of recommending novel items. To address this problem, it is desirable to consider recommendation coverage when selecting the appropriate group. This could help in simultaneously making both accurate and diverse recommendations. In this paper, we propose to focus mainly on the growing of the large search space of users’ profiles and to use an evolutionary multi-objective optimization-based recommendation system to pull up a group of profiles that maximizes both similarity with the active user and diversity between its members. In such manner, the recommendation system will provide high performances in terms of both accuracy and diversity. The experimental results on the Movielens benchmark and on a real-world insurance dataset show the efficiency of our approach in terms of accuracy and diversity compared to state-of-the-art competitors.

Résumé: Classification is a central task in machine learning and data mining. Decision tree (DT) is one of the most popular learning models in data mining. The performance of a DT in a complex decision problem depends on the efficiency of its construction. However, obtaining the optimal DT is not a straightforward process. In this paper, we propose a new evolutionary meta-heuristic optimization based approach for identifying the best settings during the construction of a DT. We designed a genetic algorithm coupled with a multi-task objective function to pull out the optimal DT with the best parameters. This objective function is based on three main factors: (1) Precision over the test samples, (2) Trust in the construction and validation of a DT using the smallest possible training set and the largest possible testing set, and (3) Simplicity in terms of the size of the generated candidate DT, and the used set of attributes. We extensively evaluate our approach on 13 benchmark datasets and a fault diagnosis dataset. The results show that it outperforms classical DT construction methods in terms of accuracy and simplicity. They also show that the proposed approach outperforms Ant-Tree-Miner (an evolutionary DT construction approach), Naive Bayes and Support Vector Machine in terms of accuracy and F-measure.

Résumé: This paper presents a new approach that avoids the over-fitting and complexity problems suffered in the construction of decision trees. Decision trees are an efficient means of building classification models, especially in industrial engineering. In their construction phase, the two main problems are choosing suitable attributes and database components. In the present work, a combination of attribute selection and data sampling is used to overcome these problems. To validate the proposed approach, several experiments are performed on 10 benchmark datasets, and the results are compared with those from classical approaches. Finally, we present an efficient application of the proposed approach in the construction of non-complex decision rules for fault diagnosis problems in rotating machines.

Résumé: Developing high performance neural networks remains a significant challenge in classification problems. This is largely due to the considerable impact of the neural network architecture on its performance. To tackle this challenge, we present a new optimization method based on particle swarm optimization and the dominating vertex set. This method aims to improve model accuracy while reducing the number of neurons and layers through an automated selection process. An experimental evaluation is conducted on several datasets from the UCI Machine Learning Repository. The obtained results show that the proposed method outperforms the multilayer perceptron in terms of classification accuracy.

Résumé: The integration of Internet of Things (IoT) technology in various domains has led to operational advancements, but it has also introduced new vulnerabilities to cybersecurity threats, as evidenced by recent widespread cyberattacks on IoT devices. Intrusion detection systems are often reactive, triggered by specific patterns or anomalies observed within the network. To address this challenge, this work proposes a proactive approach to anticipate and preemptively mitigate malicious activities, aiming to prevent potential damage before it occurs. This paper proposes an innovative intrusion prediction framework empowered by Pre-trained Large Language Models (LLMs). The framework incorporates two LLMs: a fine-tuned Bidirectional and Auto-Regressive Transformers (BART) model for predicting network traffic and a fine-tuned Bidirectional Encoder Representations from Transformers (BERT) model for evaluating the predicted traffic. By harnessing the bidirectional capabilities of BART the framework then identifies malicious packets among these predictions. Evaluated using the CICIoT2023 IoT attack dataset, our framework showcases a notable enhancement in predictive performance, attaining an impressive 98% overall accuracy, providing a powerful response to the cybersecurity challenges that confront IoT networks.

Résumé: In recent years, numerous large-scale cyberattacks have exploited Internet of Things (IoT) devices, a phenomenon that is expected to escalate with the continuing proliferation of IoT technology. Despite considerable efforts in attack detection, intrusion detection systems remain mostly reactive, responding to specific patterns or observed anomalies. This work proposes a proactive approach to anticipate and mitigate malicious activities before they cause damage. This paper proposes a novel network intrusion prediction framework that combines Large Language Models (LLMs) with Long Short Term Memory (LSTM) networks. The framework incorporates two LLMs in a feedback loop: a fine-tuned Generative Pre-trained Transformer (GPT) model for predicting network traffic and a fine-tuned Bidirectional Encoder Representations from Transformers (BERT) for evaluating the predicted traffic. The LSTM classifier model then identifies malicious packets among these predictions. Our framework, evaluated on the CICIoT2023 IoT attack dataset, demonstrates a significant improvement in predictive capabilities, achieving an overall accuracy of 98%, offering a robust solution to IoT cybersecurity challenges.

Résumé: La méthode d’ensemble des forêts aléatoires vise à concevoir un groupe d’arbres de décision construits sur la base d’un échantillonnage aléatoire sur les instances et les attributs d’apprentissage. Cette stratégie offre aux forêts aléatoires une forte capacité de généralisation. Cependant, il est primordial que pendant la construction du modèle, les arbres de décision construits soient précis (par rapport au taux de bon classement) et diversifiés (au niveau de leurs structures). Dans cet article, nous proposons une approche de construction d’une forêt aléatoire qui repose sur une optimisation collective de l’ensemble des arbres de décision du modèle. Le modèle proposé vise à : 1) trouver un bon taux de classification des arbres de décision du forêt aléatoire afin de maximiser la performance de classification du modèle ensembliste et 2) utiliser une mesure de diversité entre les arbres de décision afin d’améliorer la capacité de généralisation de la forêt aléatoire. Les analyses expérimentales effectuées sur plusieurs jeux de données montrent la supériorité de notre modèle en comparaison avec l’approche classique des forêts aléatoires ainsi que d’autres approches concurrentes

Résumé: The smart grid leverages intelligent communication technologies to modernize the electric system infrastructure. The advanced metering infrastructure (AMI) enables an intelligent bidirectional communication between the smart meters and the energy services companies. The smart grid AMI provides better customer service, energy saving, and peak reduction. However, the smart grid also, raises new vectors for cyber risks and energy frauds. In this paper, we propose a machine learning based solution to detect the fraud of energy in advanced metering infrastructure. The proposed solution employs support vector machine to identify fraudulent activity by taking advantage of the predictability of the customer consumption's profile. The performance evaluation of the proposed solution on a real electricity consumption dataset shows a high detection rate and low false positive rate compared to related works.

Résumé: In recommender systems field (RS), considering a commercial system perspective involves covering in an accurate manner most items available in the market. However, in the memory based collaborative filtering (CF),the recommendation ability is limited because of the huge size of users and items. For this reason, the clustering algorithms were employed to improve the scalability of the system by partitioning users data into clusters then performing computations on each cluster separately. We propose in this paper a recommendation approach that targets two well-known issues: the scalability problem and the recommendation diversity. Our contribution consists of two successive stages: a) K-nearest neighbor (KNN) algorithm based on the use of an adapted similarity measure. b) An adjusted neighborhood selection performed by a genetic algorithm. The approach aims to improve the quality of the neighborhood set by exploring the reduced search space obtained in the first step, to choose among them the best ones who can enhance the quality of the recommendations. The proposed algorithm was compared to baseline recommender systems and showed competitive results in terms of the diversity and the precision of the recommendations.

Résumé: Social tagging activities allow the wide set of web users, especially learners, to add free annotations on educational resources to express their interests and automatically generate folksonomies. Folksonomies have been involved in a lot of recommendations approaches. Recently, supported by semantic web technologies, the Linked Open Data (LOD) allow to set up links between entities in the web to join information in a single global data space. This paper demonstrates how structured content accessible via LOD can be leveraged to support educational resources recommender in folksonomies and overcome the limited capabilities to analyze resources information. Another limitation of resources recommendation is the content overspecialization conducting in the incapacity to recommend relevant resources diverse from the ones that learner previously knows. To address these issues, we proposed to take advantage of the richness of the open and linked data graph of DBpedia and Ant Colony Optimization (ACO) to learn users’ behavior. The basic idea is to iteratively explore the RDF data graph to produce relevant and diverse recommendations as an alternative of going through the tedious phase of calculating similarity to attain the same goal. Using ant colony optimization, our system performs a search for the appropriate paths in the LOD graph and selects the best neighbors of an active learner to provide improved recommendations. In this paper, we show that ACO also in the problem of recommendation of novel diverse educational resources by exploring LOD is able to deliver good solutions.

Résumé: In real world e-commerce applications, users express partially their preference in aim of getting back automatically valuable recommendations. The importance of that process has known an increasing development due to the high number of available products under the trade which influences negatively user choice making. The collaborative filtering approach consists of mining users/items data to model the preferences in the form of common profiles. Since ratings prediction is computed by aggregating neighbors ratings, the predictions could be calculated by performing a based similarity neighborhood selection. In this paper, we propose a weighted similarity measure as an alternative to the conventional similarity metrics used in the collaborative filtering. The proposed weights have improved novelty, diversity metrics as well as recommendation accuracy metrics. We have compared our proposed model against memory user based collaborative filtering.

Résumé: In user memory based collaborative filtering algorithm, recommendation quality depends strongly on the neighbors selection which is a high computation complexity task in large scale datasets. A common approach to overpass this limitation consists of clustering users into groups of similar profiles and restrict neighbors computation to the cluster that includes the target user. K-means is a popular clustering algorithms used widely for recommendation but initial seeds selection is still a hard complex step. In this paper a new genetic algorithm encoding is proposed as an alternative of k-means clustering. The initialization issue in the classical k-means is targeted by proposing a new formulation of the problem, to reduce the search space complexity affect as well as improving clustering quality. We have evaluated our results using different quality measures. The employed metrics include rating prediction evaluation computed using mean absolute error. Additionally, we employed both of precision and recall measures using different parameters. The obtained results have been compared against baseline techniques which proved a significant enhancement.

Résumé: Problems such as closed frequent subset mining, itemset mining, and connected tree mining can be solved in a polynomial delay. However, the problem of mining closed frequent connected subgraphs is a problem that requires an exponential time. In this paper, we present ECE-CloseSG, an algorithm for finding closed frequent unique edge label subgraphs. ECE-CloseSG uses a search space pruning and applies the strong accessibility property that allows to ignore not interesting subgraphs. In this work, graph and subgraph isomorphism problems are reduced to set inclusion and set equivalence relations.

Résumé: Fault diagnosis and condition monitoring of industrial machines have known significant progress in recent years, particularly with the introduction of pattern recognition and data-mining techniques for their development. The decision trees are among the most suitable techniques for the diagnosis and have several algorithms for their construction. Each building algorithm has its advantages and drawbacks which make the optimal choice of adapted method to the desired application difficult. In this paper we propose the diagnosis accomplishment of an industrial ventilator based on the combination vibration analysis-decision trees. For the choice of the adapted decision tree building algorithm a method based on genetic algorithms was used. Its results were commented and discussed

Résumé: Decision trees are widely used technique in data mining and classification fields. This method classifies objects following succession tests on their attributes. Its principal disadvantage is the choice of optimal model among the various existing trees types (Chaid, Cart,Id3..). Each tree has its specificities which make the choice justification difficult. In this work, decision tree choice validation is studied and the use of genetic algorithms is proposed. To pull out best tree, all models are generated and their performances measured on distinct training and validation sets. After that, various statistical tests are made. In this paper we propose the diagnosis accomplishment of an industrial ventilator(Fan) based an analysis-decision trees.