Explainability is vital in intrusion detection, as it clarifies the reasoning behind system predictions, especially when misclassifications could cause serious security breaches. By making model decisions interpretable, explainability facilitates debugging and continuous improvement. This paper advances the field by focusing on explainability and interpretability in Machine Learning (ML), particularly within Reinforcement Learning (RL). We propose a decentralized Multi-Agent Reinforcement Learning (MARL) framework with a hybrid architecture that integrates logical reasoning to design an explainable Intrusion Detection System (IDS). Logical rules provide agents with prior knowledge of their environment, improving learning efficiency and decision transparency and enhancing overall performance and reliability. The model, evaluated on the NSL-KDD benchmark dataset, achieves a 99

The non-stationarity problem poses a significant challenge in the fully decentralized cooperative multi-agent reinforcement learning setting, where the environment is inherently unstable, and agents independently learn policies that evolve over time. In this work, we address the non-stationarity challenge by introducing a novel approach integrating pruned prioritized experience replay (PPER) and collaborative learning. Unlike existing methods, PPER ensures that agents selectively retain and utilize relevant experiences, enhancing performance by pruning less significant transitions. Additionally, our approach incorporates collaborative mode, and extends knowledge sharing from training to testing phases, fostering enhanced decision-making. This framework is evaluated in the context of intrusion detection, demonstrating its applicability and potential generalizability to a variety of real-world scenarios. Experimental results on the NSL-KDD dataset demonstrate the effectiveness of our solution, achieving 98.05

Malicious actors in the Social Internet of Things (SIoT) actively exploit system vulnerabilities to manipulate trust mechanisms, leveraging the decentralized nature of interactions and the vast data generated by IoT devices. These adversaries spread deceptive information, gain unwarranted trust from users, and construct false reputations to serve harmful purposes, resulting in trust-related attacks that compromise user ratings and feedback integrity. To counter these threats, this study introduces a robust trust management framework built on blockchain technology to ensure transparency and trustworthiness. At the core of our approach is a zero-knowledge proof-based authentication layer, implemented as a smart contract, which preserves user privacy while enabling reliable verification. In addition, we develop a trust evaluation model based on a federated learning architecture tailored to the non-independent and non-identically distributed nature of data and the resource constraints of heterogeneous IoT devices. Experimental evaluations using a simulated dataset demonstrate that the proposed FedTrust model achieves up to 99.89% accuracy in detecting trust-related attacks. On-chain analysis shows that proof verification incurs approximately 390,000 gas, with costs reduced from $7.79 on Ethereum Mainnet to as low as $0.006 on Polygon, and verification latency ranging from 0.09 to 0.31 s depending on the network. These results confirm the scalability, privacy preservation, and computational efficiency of our framework, making it a viable solution for real-world SIoT deployments.

The rapid proliferation of Internet of Things (IoT) devices across sectors such as smart homes, healthcare, and industrial automation has significantly increased the attack surface, making IoT networks prime targets for cyberattacks. Traditional Intrusion Detection Systems (IDS) often struggle to balance detection accuracy with the resource constraints inherent in IoT devices. While deep learning-based IDS have shown promise, they are frequently computationally intensive, limiting their applicability in real-time, resource-constrained environments. To address these challenges, we propose a lightweight, two-stage IDS that integrates dynamic, time-aware profiling with a Convolutional Neural Network–Long Short-Term Memory (CNN–LSTM) classifier. This hybrid approach reduces computational overhead while maintaining high detection accuracy. Building upon our previous DLA-ABIDS work, this paper introduces dynamic profiling capabilities and real-time evaluation methodology. The approach is evaluated using the Ton-IoT and Bot-IoT datasets, encompassing diverse normal and malicious traffic scenarios. Experimental results demonstrate that the proposed system achieves a 98% overall accuracy with a 5% false-negative rate, indicating its potential to improve security and resilience in IoT networks.

Traffic congestion and road safety continue to be critical challenges in urban environments, especially at intersections. Cooperative Intersection Collision Avoidance (CICA) systems have emerged as promising solutions to address these challenges. CICA systems leverage Vehicle-To-Vehicle (V2V) and Vehicle-To-Infrastructure (V2I) communications to enhance safety and optimize traffic flow at intersections. Our research paper presents a novel CICA system specifically designed for unsignalized intersections. It utilizes a graph-based data model, which is particularly effective for encapsulating the complexities of Intelligent Transportation Systems (ITS) data. This system particularly focuses on rapid query processing, which is vital for effective and timely decision-making. Our method’s primary component is an algorithmic framework that analyzes these graph-based representations to identify and prevent potential traffic issues. With an emphasis on query execution time, we have thoroughly evaluated the system. This paves the way for improving road safety and reducing urban traffic congestion.

There has been a COVID-19 pandemic since December 2019, and successful medical treatment for COVID-19 patients requires rapid and accurate diagnosis. Fighting the COVID-19 pandemic requires an automated system that uses deep transfer learning to diagnose the virus on chest X-ray (CXR). CXR are frequently utilized in healthcare because they offer the potential for rapid and accurate disease diagnosis. Automated computer-aided diagnosis (CAD) systems incorporate ML or deep learning to enhance efficiency and accuracy, hence reducing future problems. Numerous AI systems based on deep learning can be employed for diagnosis; among the most widely used is the CNN, which was first developed and has demonstrated encouraging accuracy in identifying COVID-19 confirmed patients using CXR pictures. Through using X-ray images, this work will design ML and deep learning to provide faster diagnostics for COVID-19 infection. As a result, the deep transfer learning technique uses an existing model first, then applies the needed data to it again. Where a Densenet201transfer learning model was utilized, which is one of a DL techniques, as feature extraction and its combination with multilayer perceptron algorithm; these technique were applied to a data set of a National Institute Health (NIH), where several performance measures were utilized, such as precision, precision, specificity and sensitivity, as an experiment proved the efficiency of the algorithm used in terms of accuracy by 98.82%. These outcomes are encouraging when compared to other DL models that were trained on the identical dataset.

In the Social Internet of Things (SIoT), where users interact in a distributed manner, attackers exploit system vulnerabilities to manipulate trust. These attackers spread false information and services, build deceptive reputations, and gain user trust to achieve malicious goals. Such strategies, known as trust-related attacks, involve falsified ratings or manipulated feedback to artificially boost the reputation of malicious entities within the network. To counter these attacks, trust management systems play an essential role in identifying and mitigating malicious activity. Blockchain technology has revolutionized decentralized and distributed systems, providing enhanced security through various applications. Although integrating blockchain into trust management poses challenges, it significantly improves trust evaluation and strengthens the overall security framework. This work introduces a blockchain-based secure trust management system that addresses vulnerabilities through an authentication layer powered by zero-knowledge proof technology, ensuring privacy and robust validation. In addition, a trust evaluation model based on federated learning is proposed, designed to manage heterogeneous data from diverse SIoT nodes with constrained computational resources. The proposed approach is designed to detect various types of trust-related attacks, fostering trustworthy interactions within SIoT environments.

Seeking a relevant answer to a biomedical question became a daily activity not only for experts but also for patients. In this perspective, biomedical extractive Question Answering systems have witnessed a rapid progress especially with the emergence of pre-trained language models such as BERT and its biomedical variant BioBERT. Those systems aim to extract an answer to a given question from a biomedical context and rely on two principal components question processing and exact answer identification. Several pre-trained language models-based systems have been proposed and focused only on the second component. In this paper, we proposed a BioBERT-based question answering system which rests on a question expansion phase. The Latter intends to extract question terms synonyms, as expansion terms, from multiple knowledge resource MeSH and WordNet. Indeed, we used firstly BioBERT pre-training model as a representation model in the selection of relevant expansion MeSH and WordNet terms. Secondly, in the fine-tuning phase to perform the question answering task and identify the exact answer. The experimental results on BioASQ dataset highlight the interest of the BioBert-based question expansion phase.

Person re-identification (re-id) attempts to match a person from the images of different time steps. Existing deep learning approaches either use appearance or geometry features for re-id which does not provide the required robustness because of higher intra-class similarity. Existing supervised re-id approaches utilize Convolutional Neural Networks (CNNs) and identity-labeled images to train, where the person images are taken by the sensors from a horizontal view. The horizontal view exposes the privacy of the people because of their facial appearance in the image. Moreover, person re-id includes new unseen people; however, CNN does not have the ability to identify the new unseen people because of a lack of continual learning. Privacy-preserved computer vision-assisted person re-id systems can benefit from visual appearance and geometry features extracted from top-view RGB and depth input. This paper presents the privacy-preserved person top-view re-id few-shot network which uses the appearance and geometry features. The EfficientNet is used for appearance-based features from RGB input, while PointNet is used to extract the geometry features from the point cloud which is made from the RGB-D image registration. Concatenated features from EfficientNet and PointNet are fed to the two-layer Bi-LSTM network for person identification. Finally, the whole network is converted into a few-shot network to achieve continual learning by removing the output layer and joining the similarity measurement unit. This approach is based on CNN and fine-tunes a TVPR/2 dataset acquired by using a top-view arrangement that is publicly available. The experimental results on TVPR/2 and GODPR datasets show that the proposed re-id network outperforms other state-of-the-art networks.

Recently, researchers and clinicians have been searching for new technologies to slow down or stop COVID-19 pandemic. The utility of Data Mining (DM) algorithms to suggests new opportunities to combat the spread of the new Coronavirus. This paper suggests a comparative study on data mining approaches to predict COVID19. We used common classification algorithms like the Support Vector Machines, Random Forest, Logistic Regression, K-Nearest Neighbor and Artificial Neural Network with Python simulation to compare it in metrics accuracy, recall, precision and AUC; results showed that Random Forest model had a 98.43% accuracy – which is a higher accuracy than many other previous studies known COVID-19 data mining algorithms.

Intelligent systems knowledge bases are grounded on ontologies and their associated rules. As intelligent systems act in dynamic and unpredictable environments, their knowledge bases need to be updated. Inevitable anomalies may arise and disturb the systems' performance. Thus, rule bases verification became an indispensable task. In this paper, we propose an approach called TARO (Treating Anomalies in Rule bases associated to Ontologies). Our solution is structured in two steps: first, we extract dependency relationships between rules, then we use these relationships to identify and resolve anomalies. We developed a working prototype of our proposal which we named RB-Verif. Our approach's effectiveness was demonstrated by the experiments carried out.

The Social Internet of Things (Social IoT) introduces a fresh approach to promote the usability of IoT networks and enhance service discovery by incorporating social contexts. However, this approach encounters various challenges that impact its performance and reliability. One of the most prominent challenges is trust, specifically trust-related attacks, where certain users engage in malicious behaviors and launch attacks to spread harmful services. To ensure a trustworthy experience for end-users and prevent such attacks in real-time, it is highly significant to incorporate a trust management mechanism within the Social IoT network. To address this challenge, we propose a novel trust management mechanism that leverages blockchain technology. By integrating this technology, we aim to prevent trust-related attacks and create a secure environment. Additionally, we introduce a new consensus protocol for the blockchain called Spark-based Proof of Trust-related Attacks (SPoTA). This protocol is designed to process stream transactions in real-time using Apache Spark, a distributed stream processing engine. To implement SPoTA, we have developed a new classifier utilizing Spark Libraries. This classifier is capable of accurately categorizing transactions as either malicious or secure. As new transaction streams are read, the classifier is employed to classify and assign a label to each stream. This label assists the SPoTA protocol in making informed decisions regarding the validation or rejection of transactions. Our research findings demonstrate the effectiveness of our classifier in predicting malicious transactions, outstripping our previous works and other approaches reported in the literature. Additionally, our new protocol exhibits improved transaction processing times.

Intrusion detection systems (IDSs) play a crucial role in network security, as the need for secure networks continues to grow. However, traditional IDSs are not able to accurately and efficiently detect attacks due to the vast amount of data generated in a network and the emergence of new types of attacks every day. To address this challenge, researchers have recently turned to reinforcement learning (RL) techniques to enhance IDSs’ accuracy and efficiency. RL is a powerful machine learning approach that can learn from experience and adapt to changing environments, making it a promising choice for IDS applications. In this paper, we present a comprehensive survey of RL-based IDSs. We provide an overview of various RL algorithms applied to IDSs and discuss the different types of features used to represent network traffic data. Additionally, we highlight the challenges and future directions in the field to improve the performance of RL-based IDSs.

Document indexing phase plays a significant role in text mining applications such as text document classification. The common indexing paradigm is based on terms frequency in documents known as Bag Of Words (BOW)-based representation approach. However, such classical approach suffers from ambiguity and disparity of words. In addition, traditional term weighting schemes, such as TF-IDF, exploit only the statistical information of terms in documents. To overcome these problems, we have been interested in biomedical semantic document indexing using concepts extracted from the knowledge resource MeSH. Thus, we have focused first on a disambiguation method to identify the adequate senses of ambiguous MeSH concepts and we have considered four representation enrichment strategies to identify the best appropriate representatives of the adequate sense in the textual entities representation. Second, we propose to introduce a semantic weighting scheme that quantifies MeSH concept’s importance in documents through their occurrence frequency and semantic similarities with unambiguous MeSH concepts. Our contribution lies particularly in the in-depth experimental study of the performance of these methods and precisely the impact of the semantic weighting scheme on the performance. To do that, three benchmark datasets TREC 2004 genomics, BioCreative II and OHSUMED were used.

The Social Internet of Things (SIoT) facilitates seamless interactions between IoT devices, providing users with quick and convenient services. However, this domain is vulnerable to manipulation by malicious nodes that issue false recommendations and services to inflate their reputation, leading to trust-related attacks. Developing trust models to detect these attacks in each interaction is challenging due to the complexity of the patterns and features required for accurate prediction. Furthermore, trust metrics are not consistently updated for each node, resulting in inefficiencies and unnecessary resource consumption. To address these challenges, we propose a system that analyzes the context of the current interaction and incorporates temporal factors to monitor node behavior. Our approach employs a decentralized system based on blockchain and IPFS storage, reducing costs and making the process of trust evaluation more efficient and practical for real-time scenarios. This method enhances the detection of trust-related attacks while optimizing resource allocation and execution time.

Integrating the Internet of Things (IoT) with Social Networks (SN) has given rise to a new paradigm called Social IoT, which allows users and objects to establish social relationships. Nonetheless, trust issues such as attacks have emerged. These attacks can influence service discovery results. A trust management mechanism has become a major challenge in the Social IoT to prevent these attacks and ensure qualified services. A few studies have addressed trust management issues, especially those that prevent trust attacks in Social IoT environments. However, most studies have been dedicated to detect offline attacks with or without specifying the type of attack performed. These works will not be able to prevent attacks by aborting transactions between users because their primary purpose is to detect an offline attack. In addition, they do not consider security properties. This research paper aims to provide a detailed survey on trust management mechanism to handle trust attacks in Social IoT. In this research paper, we compared the techniques and technologies whose common point is attack prevention and demonstrated that blockchain technology can play a key role in developing a trust management mechanism that can prevent trust attacks while maintaining security properties. Then, we proposed combining the Apache Spark Framework with blockchain technology to provide real-time attack prevention. This combination can assist in creating upgraded trust management mechanisms in Social IoT environments. These mechanisms aim to prevent attacks in real-time through considering the security properties. Lack of survey papers in the area of trust attack prevention in real-time stands for an important motivational factor for writing this paper. The current research paper highlights the potential of the blockchain technology and Apache Spark in terms of developing an upgraded trust management able to prevent trust attacks in real-time.This paper provides a comprehensive survey on trust management mechanisms and approaches to handle trust attacks in Social IoT. Lack of such papers increases the significance of this paper. It also offers potential future research directions in terms of real-time trust attack prevention.

The growing complexity of security threats and the pervasive prevalence of cyberattacks have become more apparent in the present era, and the advent of big data, characterized by its distinctive features, has introduced layers of complexity to security tasks. Intrusion Detection Systems (IDSs) constitute a crucial line of defense, but their adaptation to the realm of big data is imperative. While traditional Machine Learning (ML)-based IDSs have been pivotal in detecting malicious patterns, they are often incapable to keep pace with the demands of expansive big data networks. This paper proposes a novel decentralized Multi-Agent Reinforcement Learning (MARL)-based IDS designed to address the specific challenges posed by big data. Our solution employs decentralized cooperative MARL, securing communicative channels throughout the detection process and concurrent data preprocessing which significantly reduces the overall processing time. Furthermore, the integration of Cloud computing and Big Data streaming techniques further facilitates real-time intrusion detection as cloud’s resources allow rapid pre-process and analyse of massive data streams using powerful clusters. Likewise, Big Data streaming techniques ensure that potential intrusions are identified and addressed as they occur. Experimental results, conducted on the widely recognized NSLKDD benchmark dataset, demonstrate the superiority of our solution over other state-of-the-art approaches for big data networks, achieving an accuracy rate of 97.44%.

Query expansion represents a crucial task and a decisive step in many information retrieval applications such as document retrieval and question answering. Its success has a major impact on the performance of the subsequent steps. In this paper, we present a global corpus based-query expansion method. This method relies on the one hand on the WordNet knowledge resource to expand documents. On the other hand, it rests on pre-trained embedding models Word2Vec, GloVe and BERT to identify and select the relevant expansion embedding terms. Our main motivation is to combine the knowledge contained in structured and unstructured semantic resources by exploiting the advantage of introducing deep learning which seeks to bring out the semantic relationships buried in large corpus of documents. The experimental results in the context of document retrieval process showed that our method performs well.

Textual documents representation is considered as a crucial phase in many text mining tasks. It intends generally to capture the semantic information of the whole document into a representation vector, which could be further utilized in such tasks. In this paper, we have proposed a deep semantic representation method which is based on two types of features. The first is derived from the biomedical widely used Structured Semantic Resource MeSH whereas the second is generated from a deep learning phase. At this end, we chose to combine the two models Word2Vec and Convolutional Neural Network which allow bringing out the semantic relationships existing in large and complex textual documents at high abstraction levels. To evaluate this method, we involved it into a document classification process. The results of experiments, carried out on a sub-set of the OHSUMED collection, show that our method performs well.

Ensuring the security of computer networks is of utmost importance, and intrusion detection plays a vital role in safeguarding these systems. Traditional intrusion detection systems (IDSs) often suffer from drawbacks like reliance on outdated rules and centralized architectures, limiting their performance in the face of evolving threats and large-scale data networks. To address these challenges, we present an advanced anomaly detection-based IDS that utilizes a decentralized communicative multi-agent reinforcement learning (MARL). In our approach, multiple reinforcement learning agents collaborate in intrusion detection, effectively mitigating the non-stationarity problem and introducing a specialized secure communication method. We further enhance the learning process by incorporating external knowledge. Our approach is evaluated through extensive experiments conducted on the benchmark NSL Knowledge Discovery and Data Mining dataset. These experiments encompass diverse scenarios, involving varying numbers of agents to prove scalability feature. The results underscore the effectiveness of our method, which surpasses the performance of existing state-of-the-art solutions based on MARL, achieving a high accuracy rate of 97.80%.