The proliferation of Internet of Things (IoT) devices has created unprecedented cybersecurity challenges, with these devices becoming prime targets for attacks due to their computational constraints and valuable data repositories. While Intrusion Detection Systems (IDSs) serve as critical defensive mechanisms, traditional approaches struggle with the scale, heterogeneity, and privacy requirements of IoT ecosystems. This paper presents a comprehensive analysis of Machine Learning (ML) techniques for IoT-focused IDSs, progressing from conventional ML algorithms through Deep Learning (DL) architectures to privacy-preserving Federated Learning (FL) approaches. This paper compare these methodologies across key performance metrics including detection accuracy, privacy preservation, and computational efficiency. The analysis reveals that while FL offers promising solutions for privacy-sensitive IoT environments, significant challenges remain, including communication overhead, model personalization, and vulnerability to poisoning attacks. The paper categorize these challenges and synthesize potential solutions proposed in recent literature, providing a roadmap for future research directions in developing robust, privacy-preserving intrusion detection mechanisms for IoT networks.

Multi-access Edge Computing (MEC) is an emerging paradigm designed to provide storage, computing and communication capabilities in the proximity of end-user devices. This approach facilitates the deployment of real-time on mobile devices with limited capabilities. To realize the MEC goals, it is essential to effectively manage and offload computing tasks to both edge and cloud-based resources. However, the dynamic nature, uncertainty and mobility within edge computing environments pose significant challenges to resource management. Furthermore, the inherent software and hardware heterogeneity, coupled with the distributed nature of architecture, complicates the development of efficient task offloading strategies that can adeptly manage resources across both edge and cloud platforms. In this paper, we propose a cluster-based task edge orchestrator, where edge servers are grouped based on service demands, resource ability and other factors to improve the overall service. Our proposed method leverages the K-Medoids clustering algorithm to dynamically form clusters of suitable edge servers for offloading computing tasks with minimum response time. To validate our proposed solution, we have orchestrated a comprehensive series of tests using EdgeCloudSim. Results show that our approach outperforms its competitor in terms of average service time by around 8 %.

AI analytics enables autonomous cars to detect and recognize objects, such as other vehicles, pedestrians, traffic signs, and obsta- cles, in real-time. Deep learning models, notably the You Only Look Once (YOLO) model, have demonstrated accuracy and speed in obstacle avoidance. However, current datasets are limited, lacking diversity and labeling, hindering their ability to represent real- world scenarios accurately. Besides, previous studies have focused extensively on specific object classes, such as pedestrians and vehicles, often neglecting other objects like bikes and road signs. To address this, we introduce a novel dataset tailored for AV envi- ronments, encompassing various road object types under different conditions. Our innovative methodology relies on self-supervised learning using the late YOLO version to improve model robustness with limited labeled data and AI-driven adaptive model opti- mization based on real-time feedback. We evaluate three YOLO architectures—YOLOv5, YOLOv7, and YOLOv8—customized for AV object detection. Our assessment covers everyday AV objects such as cars, pedestrians, bicycles, and road signs, empha- sizing early detection. We employ the VSim-AV simulator dataset to ensure robust evaluation, augmented with preprocessing techniques to optimize data quality and model generalization. The study reveals that YOLOv5 and YOLOv8 outperform YOLOv7 regarding precision and recall across various object classes, with YOLOv5 leading at 1.3 ms/image and YOLOv8 at 3.3 ms/image. The mean average precision was 0.94 for YOLOv5, 0.441 for YOLOv7, and 0.927 for YOLOv8, highlighting the limitations in current literature and challenges in YOLO model performance.

Speaking is the most natural way of communicating with others. Speech recognition technology aims to enable machines to comprehend and act upon human speech. In the field of education, speech recognition has proven to be a valuable tool for automatic language correction, particularly for Quran recitation. Correct recitation of the Quran is crucial for Muslims, and traditionally, it requires an expert "gari" to identify and correct mistakes. While this method is effective, it’s also time-consuming. Quranic speech recognition apps can help students with their recitation, but they require a robust and accurate speech recognition model. While recent models for Arabic and non-Arabic speech recognition have been successful, research on Holy Quran speech recognition is still in its early stages. As such, this study aims to propose an end-to-end deep learning model for Holy Quran speech recognition using the ESPnet toolkit.

Computational task offloading facilitates real-time applications on constrained mobile devices that require a large amount of processing resources, high storage capacity and battery power. Mobile Edge Computing (MEC) is a computing paradigm that shifts computing resources closer to the user at the network’s edges. Heavy tasks are then offloaded to edge nodes, thereby reducing the computations required on the mobile side. However, offloading computational tasks may result in additional energy consumption and delays, due to network congestion and time spent in server queues. Thus, to minimize completion time and energy consumption, it is essential to optimize offloading decisions, while also considering the financial costs. In this paper, we propose an Offloading Decision-Making Framework based on the Binary Cuckoo Search Algorithm for Mobile Edge Computing (ODM-BCSA). We formulated an offloading problem as a mixed-integer optimization problem to minimize time, energy, and payment costs. We resolved the problem of resource allocation using the Binary Cuckoo Search Algorithm (BCSA). The simulation results revealed that offloading decisions depend on multiple parameters, including the number of mobile devices being handled by the edge server, the bandwidth, and the number of tasks. Decisions made also depended on the priority assigned to each objective. Finally, we compared the ODM-BCSA against a brute force search, proving that the ODM-BCSA is more efficient and greatly minimizes execution time when a huge number of mobile devices are involved (by 99.9 %).

Managing and storing big data is non-trivial for traditional relational databases (RDBMS). Therefore, the NoSQL (Not Only SQL) database management system emerged. It is ca-pable of handling the vast amount and the heterogeneity of data. In this research, we are interested in one of its trending types, the graph database, namely, the Directed Property Graph (DPG). This type of database is powerful in dealing with complex relationships ($\mathrm{e}.\mathrm{g}$., social networks). However, its sen-sitive and private data must be protected against unauthorized access. This research proposes a security model that aims at exploiting and combining the benefits of Access Control, View-Based, and Query-Rewriting approaches. This is a novel combination for securing DPG.

Asthma is a common respiratory disease affected by different biosignals and environmental triggers. Early prediction of asthma attacks is crucial to saving patient lives. Several machine learning models have been designed to predict asthma attacks. However, few researchers have exploited biosignals and environmental triggers to build asthma attack prediction models. Additionally, little attention has been devoted to feature selection algorithms and the variation of machine learning models. This study develops an asthma attack prediction model by testing different machine learning classifiers. The dataset used includes two main parts: the biosignals dataset, which was recorded daily from 21 volunteers for three months, and the environmental dataset, which is available online. The machine learning classifiers used in the study are decision tree, gradient boosting models, logistic regression, random forest, and support-vector machine. Each classifier was grid searched to find the best value for the primary hyperparameters; then, we used five-fold cross-validation to train each model. Results show that the gradient boost model outperforms the other classifiers when trained with 0.5 for the depth parameter and 9 for the sub-sample parameter. The prediction of the testing set produces 97.2% accuracy and 97.1 % recall.

As fraud and cybercrime become more frequent and sophisticated, preventing users from being exposed to risk is a significant challenge for the scientific community. Current banking systems use One Time Password (OTP) sent to the mobile phone to prevent user impersonation attacks. However, it could be compromised by man-in-the-middle attacks. It also presents usability and unavailability problems if the SMS service is inaccessible, especially when traveling or in case of a damaged or stolen phone. Using Machine Learning (ML) techniques to detect and prevent fraudulent actions allows financial organizations to stay one step ahead of fraudsters, whose scale and sophistication continue to grow. Specifically, ML systems can be used to identify additional authentication factors to make the system more robust against attacks. To prevent identity theft attacks and minimize fraudulent transactions, we propose in this work a comprehensive design that combines adaptive multi-factor authentication and adaptive fraud detection for banking systems based on the use of ML technology.

Nowadays, information retrieval systems face significant challenges in providing users with accurate information due to the enormous growth of information. To better reformulate the query and narrow its results, semantic query expansion techniques add semantically related terms to the original query. However, semantic query expansion for Arabic queries is still a challenge due to the lack of rich semantic sources. Most of the existing solutions rely on using either English sources or specific-domain Arabic ontologies. Using English sources requires a translation phase which may lead to query drift, thus providing unrelated expansion terms. In this paper, we provide an overview of the query expansion approaches. Besides, we propose a hybrid comprehensive reference framework for Arabic semantic query expansion that overcomes the lack of Arabic semantic sources by using rich English ontologies to complement the limited Arabic sources (Arabic Wordnet) currently available. It ensures the Arabic-English translation process using a customized machine learning translation model to avoid query drifting. It also transforms natural language to SPARQL (an ontology query language) to easily query English sources (e.g., DBpedia). For enhanced accuracy, it provides an optimization module where meta-heuristics can be used for pertinent terms selection. This work represents a step forward in combining English sources and AI to design a practical Arabic semantic expansion.

Collaborative editing applications for cloud environment play an important role in many fields and communities since they allow users to communicate and collaborate using their mobile devices. Mobile devices are usually cloned in the cloud to minimize computation cost and energy consumption. Moreover, these mobile and cloud interactions lead to online and offline switching in easy and continuous ways in order to edit shared multimedia documents. However, the main concern of these applications is still maintaining consistent and secure copies of the shared documents with low latency and high local responsiveness. Indeed, appropriate access control models are needed to preserve the features of collaborative editing applications when combining mobile and cloud environments. In this paper, we present a study on existing cloud-based collaborative editors and the current state of the art of access control models used in the collaborative edition context. This study has raised a series of shortcomings that have enabled us to sketch a new access model for deploying securely mobile collaborative editing applications in the cloud.

Collaborative editors are one of the most popular collaborative tools. They are being widely used thanks to the success of data sharing platforms where in most of the cases, data is shared with the intent to be edited simultaneously by many users who are distributed and dispersed across the globe. Keeping shared data synchronized is resource-intensive. With the emergence of collaborative editors over mobile devices, the challenges to meet increasing communication and computation are more and more noticeable. Indeed, the centralized cloud-based architecture incurs high delays which prevents users from seeing shared data updates in real-time fashion. In this paper, we give an overview on existing cloud-based works and propose a new edge-based architecture for collaborative editors. Mobiles that are managed by the same edge are cloned to offload resource-intensive tasks to the edge node, whereas only lightweight edition components are handled locally. This provides a more effective solution to manage concurrency and collaboration on mobile devices.

The mobile phone technologies are becoming pervasive in recent years. These items such as IPhones, IPad and Androïds are very attractive since they provide relatively good resources for a mobile device. Several works aim at integrating desktop applications in these tools to make them closer