Federated learning (FL) enables decentralized model training without sharing raw data, but its use in wireless sensor networks (WSNs) for natural disaster management remains underexplored. In this paper, we address this gap by proposing a data-aware clustered FL system tailored for disaster scenarios. We introduce the Data-Aware Disk Covering Problem (DA-DCP), a clustering method that leverages central knowledge of data distributions to form balanced clusters. These clusters serve as FL agents, improving both clustering efficiency and model convergence under heterogeneous data conditions. The simulation results highlight the advantages of DA-DCP in accelerating learning and improving robustness for disaster response.

Our innovative federated learning approach addresses the evolving landscape of collaborative machine learning by strategically sharing $80 \%$ of the dataset among decentralized devices. Utilizing TensorFlow Federated and integrating privacy-preserving techniques like differential privacy, our framework seeks to harmonize enhanced model accuracy with robust privacy preservation (20%). Through extensive experiments, we showcase the significant improvement in model accuracy (98.53%) compared to traditional federated learning (97.55%). Analyzing trade-offs between accuracy and privacy preservation, we offer insights into the impact of varied data partitioning ratios and privacy-preserving parameters. This contribution presents an effective methodology for privacy-conscious collaborative model training, unveiling opportunities to optimize federated learning processes and emphasizing the delicate balance between model performance and privacy in collaborative machine learning environments.

Federated Learning (FL) is a machine learning (ML) approach that allows a model to be trained across multiple decentralized devices holding local data samples without exchanging them. In the realm of Wireless Sensor Networks (WSNs), FL has not attracted much attention given that FL is typically meant to train models on data collected by much fewer and decently more powerful devices. However, given the potential of WSNs to collect diverse data in hazardous regions, we aim to explore how to employ FL to collect data in an area of interest where we have a natural disaster. In this paper, we introduce a novel task with regards to FL in the context of WSN for natural disaster management. Given a region where wireless sensors are deployed and data samples are distributed, we aim to cluster the sensors so that each cluster can be treated as a FL agent in a way that accelerates the process of FL.

Abstract Artificial intelligence (AI) techniques implemented at a large scale in intelligent transport systems (ITS), have considerably enhanced the vehicles' autonomous behaviour in making independent decisions about cyber threats, attacks, and faults. While, AI techniques are based on data sharing among the vehicles, it is important to note that sensitive data cannot be shared. Thus, federated learning (FL) has been implemented to protect privacy in vehicles. On the other hand, the integrity of data and the safety of aggregation are ensured by using blockchain technology. This paper applied classification approaches to VANET and ITS cyber‐threats detection at the vehicle. Subsequently, by using blockchain and by applying an aggregation strategy to different models, models from the previous step were uploaded in a smart contract. Lastly, we returned the updated models to the vehicles. Furthermore, we conducted an experimental study to measure the effectiveness of the proposed prototype. In this paper, the VeReMi data set was distributed in a balanced manner into five parts in the experimental study. Thus, classification techniques were executed by each vehicle separately, and models were generated. Upon the aggregation of the models in blockchain, they were returned to the vehicles. Lastly, the vehicles updated their decision functions and accessed the precision and accuracy of cyber‐threat detection. The results indicated that the precision and accuracy decreased by 7.1% on average with comparable F1‐score and recall. Our solution ensures the privacy preservation of vehicles whereas blockchain guarantees the safety of aggregation technique and low gas consumption.

Arabic text recognition is a difficult task due to the cursive nature of the Arabic writing system, the different forms of Arabic characters in words, the large number of ligatures, and many other challenges. Deep Learning models have made significant progress in many fields, especially in the field of Optical Characters Recognition (OCR). This article presents a model capable of recognizing handwritten Arabic characters based on deep learning. The proposed model uses Convolutional Neural Networks (CNNs) to divide the 28 Arabic characters into subclasses to improve the classification phase in OCR. The model was tested on the Handwritten Arabic Characters Database (HACDB) dataset and it gave 98% of recognition rates.

The purpose of the research on Vehicular Ad hoc NETwork (VANET) is to improve road safety, provide passenger comfort, and prevent accidents. Messages sent through VANETs are vulnerable to a variety of misbehaviors. Traditional techniques, like encryption, are ineffective since there is no purpose in being impervious to insider misbehavior. In this study, we propose an automatic learning method for detecting the misbehaving message deliver through a vehicle in VANETs using machine learning and deep learning techniques. To evaluate the performance of these techniques, we used the first publicly available Vehicular Misbehavior Dataset VeReMi. In this paper, we will compare the performance of ML and DL in VANET for detecting and classifying misbehavior messages.

The safe deployment of Internet of Things (IoT) devices has emerged as one of the most pressing issues in computer science. The Lossless Low Power Network (RPL) routing protocol is an excellent choice for Internet of Things routing. The ubiquity, connectivity, and low processing capabilities of IoT network devices identify them. These features, along with their exponential growth in recent years (more than 60 billion devices linked to the Internet by the end of 2021), have led in a surge in IoT-based cyber-attacks. To fight against these cyber-attacks, an intrusion detection system (IDS) is required to secure the privacy, availability, and performance of the IoT network. Unfortunately, most public data sets related to IDS, such as UNSW NB15 and KDD CUP99, are incompatible with the IoT network’s unique environment. To solve these issues, we created a data collection framework that includes the recording of network traffic from its unique environment to IoT device needs. The examined dataset’s advantages are tailored to 6LoWPAN/RPL and CoAP, the most widely used protocols for IoT network deployment.

Support Vector Machines (SVM) is a statistical classification approach which has been successfully applied to various types of problems. However, it has remained largely unexplored for Arabic recognition. SVMs are originally designed for binary classification problems. For multi-class problems, several methods used a combination of binary SVMs while some others solved the problem in one step. This paper introduces an evaluation of five SVM methods for the Arabic characters recognition problem; three are based on binary classifiers, and two considers all classes at once. The selected algorithms are compared in terms of training time, testing time and accuracy. Experiments conducted using the Arabic Printed Text Image Database-Multi-Font(APTID/MF ) showed that the “one-against-one method” is the robust, fast and produces a very good score rate at a reasonable time.

This paper describes a new technique for clustering data based on their trend characteristics. The technique that we propose proceed by incorporating a new distance based on qualitative trend analysis into Mean shift clustering algorithm. Mean shift clustering is a powerful non-parametric technique that does not require prior knowledge of the number of clusters and does not constrain the shape of the clusters. Trend analysis is a data-driven semi-quantitative technique that has been used for process monitoring and fault detection and diagnosis. The performances of our approach are assesed through synthetic banana shaped data. Unsupervised clustering is then applied for intelligent decision-making process specifically for fault diagnosis on Tennessee Easteman Process (TEP) challenge.