Earthquakes are among the most devastating natural disasters, and accurately forecasting their magnitude is critical for reducing their impact on life and property. In this study, the effectiveness of advanced machine learning techniques for improving earthquake magnitude prediction is examined using comprehensive seismic data. A hybrid methodology was developed that integrates machine learning models with metaheuristic feature selection methods to enhance accuracy and robustness. Feature selection was performed using Genetic Algorithm, Particle Swarm Optimization, and Simulated Annealing, while ten machine learning models were implemented — ranging from Linear Regression and Decision Trees to Gradient Boosting, XGBoost, LightGBM, and Long Short-Term Memory (LSTM) networks. An ensemble approach was proposed, incorporating Voting and Stacking Regressors to leverage model diversity for improved performance. The Stacking Regressor outperformed all other models with a minimum RMSE of 1.75, MAE of 1.35, and an R-squared value of 0.94. These results demonstrate the potential of ensemble techniques for accurate and reliable earthquake magnitude forecasting, which may contribute significantly to disaster preparedness and risk mitigation.

Assessing English-language teaching from questionnaire data requires a representation that preserves nonlinear response patterns while retaining the relative importance of pedagogical constructs. This study presents a pedagogically guided wavelet-enhanced autoencoder for teaching-effectiveness assessment. A weighted reconstruction objective is defined, with 25 questionnaire indicators allocated to 7 educational domains, and the associated domain weights are included in the reconstruction objective. In the encoder, Morlet and Ricker wavelet activations are assessed and learned bottleneck representations are analysed with a deep neural network and radial basis function predictors. The evaluation is based on 1000 analytic records and with a ten rotation training/validation/testing protocol of 80:10:10 features. The best DNN configuration, with 256 hidden units, had an AUC of 0.970, an accuracy of 0.953, a sensitivity of 0.948, and a specificity of 0.956. The explicit link between pedagogical priorities and representation learning is provided by the framework, but evidence of the current work is based on questionnaire-derived data collected in one institutional context, which needs to be validated for use outside the current context.

Background: Arabic Sign Language (ArSL) recognition remains limited in terms of technological development, compared to American Sign Language (ASL). This disparity restricts communication accessibility for individuals with hearing impairments in Arabic-speaking regions, in offline environments with limited computational resources.Objective: This study aimed to develop a robust offline recognition system for ArSL by integrating Principal Component Analysis (PCA) for dimensionality reduction, Scale-Invariant Feature Transform (SIFT) for feature extraction, and Convolutional Neural Networks (CNNs) for gesture classification.Material and Methods: This experimental, quantitative research used a curated dataset of ArSL gestures, obtained from Kaggle. Preprocessing involved normalization, contrast enhancement, and noise reduction. SIFT was used to extract invariant features, while PCA reduced computational complexity. CNN architectures were trained to recognize gestures, assessed using accuracy, precision, recall, F1-score, loss, confusion matrix, and Receiver Operating Characteristic (ROC) curve.Results: The system achieved an accuracy of 86.64%, surpassing conventional models, such as SIFT combined with Support Vector Machines (SIFT+SVM) (84.45%). The integration of PCA and SIFT enhanced recognition efficiency and reduced model complexity. Deep learning methods showed superior adaptability and precision across gesture types. Conclusion: This study presents a robust offline ArSL recognition system that enhances communication, education, and social participation for individuals with hearing impairments in Arabic-speaking regions.

Global Navigation Satellite Systems (GNSS) provide essential positioning services for unmanned aerial vehicles (UAVs), yet they are highly vulnerable to adversarial attacks such as GPS spoofing and GNSS meaconing. These attacks can mislead navigation systems, disrupt mission-critical operations, and compromise the safety of UAV-based IoT applications. Traditional detection techniques often focus on single-attack scenarios or rely on handcrafted signal features, limiting their adaptability to diverse and evolving threats. To address this challenge, we propose a multi-task learning (MTL) framework that jointly detects GPS spoofing and GNSS meaconing while simultaneously estimating confidence levels in navigation integrity. The framework leverages shared feature representations extracted from raw GNSS observables, including signal-to-noise ratios, pseudorange residuals, Doppler shifts, and carrier phase consistency, which are processed through a shared backbone network with task-specific output layers. By optimizing for multiple objectives concurrently, the model captures both commonalities and unique patterns of spoofing and meaconing attacks. Experimental evaluation on a hybrid dataset combining simulated UAV flight traces with real GNSS spoofing and meaconing scenarios demonstrates that the proposed approach achieves 99.87% accuracy, 99.7% precision, and an AUC of 0.996, significantly outperforming conventional single-task baselines. These results confirm the effectiveness of MTL in enhancing the robustness, efficiency, and scalability of UAV IoT security against GNSS threats.

The development of the resource allocation process is based on essential factors, such as population size, which can be considered one of the most critical ones. Estimating the population size for a particular region, especially in developing countries, indicates the government’s authority to invest and utilize the resources efficiently. As a result, forecasting population size is one of the most essential requirements that guide the entire development process and, as a result, affects human well-being. This paper presents a novel technique for census seven prediction of a specific region of interest in the context of satellite images. The Multidimensional Binary Layered Model (MDBL) is a representation technique founded on identifying a given region of interest in terms of the geographical nature of the local surrounding regions. The MDBL is then employed to generate an "effective" classifier using different machine learning techniques such as the Convolutional Neural Network (CNN), k-nearest Neighbor (k-NN), and Support Vector Machine (SVM). The Black Widow Optimization (BWO) method has been used to maximize performance and provide the best machine learning model. In order to produce the best prediction model, this optimization method ensures the models are tuned with the most appropriate hyperparameters. To this end, Al-Karak City, in Jordan, has been used as a case study to evaluate the proposed model. The evaluation of the classifiers has been reported in terms of the accuracy and the Area Under Curve (AUC) measurements. The CNN achieved the “best” overall AUC result of 88

Phishing is a cyber threat where attackers create deceptive websites or emails to exploit individuals for fraudulent purposes such as providing sensitive information like usernames, passwords, or financial details. Hence, detecting phishing emails is of paramount importance to protect sensitive information, preventing financial losses, protecting against identity theft, and maintaining trust. This study employs machine learning techniques to classify and categorize websites that are used for phishing attacks. To achieve this, we proposed a classification model for the detection of phishing domains based on classification rules. These rules were formulated based on analyzing and distributing them into three classifiers: domain name features, search engine opinion (SEO) information of the domain, and security scans by security vendors. A total of eleven classification rules were formulated which were distributed as follows: six rules for the first classifier, four rules for the second classifier, and one rule for the third classifier. Each classifier feeds its results to the main classifier, which then evaluates the received results based on its rules. Domain names from 222 confirmed phishing URLs were used to test the suggested model. The model demonstrated exceptional performance, achieving an impressive accuracy rate of 99.1% during testing.

The rapidly expanding universe of Internet of Things (IoT) networks has led to greater security risks due to the proliferation of sophisticated threats such as scanning activities, where adversaries use scanning to identify and exploit vulnerabilities in target IoT devices, which often have limited computing power and storage, making them attractive targets; to address this escalating challenge, this research proposes a comprehensive ensemble learning methodology specifically tailored for robust and timely detection of scanning activities in IoT networks, integrating a diverse set of base classifiers including Support Vector Machine (SVM), Neural Network (NN), Decision Trees (DT), and Linear Regression (LR), with predictions from the base models combined using a Linear Regression meta-classifier into a stacked ensemble model; extensive comparative evaluation on an IoT network dataset demonstrates the proposed ensemble technique achieves significantly higher accuracy, detection rate, F-measure, precision, and Area Under ROC Curve (AUC) versus individual traditional machine learning models, and remarkably, exhibits consistent performance across different cross-validation folds, highlighting its inherent stability and validity, suggesting it generalizes well without overfitting; in summary, this research puts forward an efficient and highly effective ensemble learning solution for the timely and precise detection of scanning attacks against IoT networks, serving as a vital step towards strengthening security against escalating threats..

Forensics investigators must often go deep into storage media in order to unearth long-lost files. Due to its prevalence in criminal activity, the Windows operating system is becoming an integral part of residual forensics. This study investigates how effective computer forensic data recovery tools are at rescuing evidence from a wide range of storage media. Our goal is to determine how much information can be retrieved by the computer image carver using the free tools FTK, EnCase, and Recuva. The results showed that the forensic tools were effective in their various recovery capacities, however this was only proved in highly specialized environments. In this term paper, we'll explain what data loss is, what causes it, how to prevent or reduce data loss, and what strategies work best for recovering and maintaining data.

Current cutting-edge solutions to the spectrum shortage problem are unable to meet the growing demand for a limited spectrum. A key dimension beyond state-of-the-art solutions is to exploit the free spectrum more effectively. Although various schemes have been proposed for trading spectrum, few studies have focused on optimal admission of spectrum requests for maximizing service providers' (SP's) profit. Thus, this timely study presents a novel intelligent admission scheme for spectrum requests from the perspective of a non-cooperative game, in which the information of all participants (customers and providers) is incomplete to others, and each player wishes to maximize its benefit. The proposed control admission policy may evict clients in-service to release spectrum for serving certain, e.g., wealthy clients. Evicted clients are compensated using a dynamic strategy that adopts greedy game theory to capture the conflict of interest between SP and evicted users. Simulation experiment results validate and demonstrate the feasibility and efficiency of the proposed scheme, compared to a benchmark reinforcement learning approach and another widely used scheme for admission and eviction control of cognitive radio users.

With the great widespread networking and Software-Defined Network (SDN) solutions, software-defined networks have become the target of many different attacks and security threats. Software-defined networks are frequently exposed to denial-of-service attacks and distributed denial-of-service (DDoS), which may harm the controller or switch of SDN.. Consequently, the services offered by this network can be negatively affected.In this research, an experimental work was conducted to detect a DDoS Flooding attack. The features were extracted from a dataset to understand the behavior of the SDN and measure its performance in case of normally operating or when it is subjected to a DDoS attack.The performance of SDN was evaluated using several machine learning classifiers. Three classifiers are used in our experiments: Random forest (RF), Support vector machine (SVM), and Naive Bayes (NB).The results showed the superiority of the RF classifier over other classifiers with a detection accuracy of 98.89%.

Plant diseases represent one of the critical issues which lead to a major decrease in the quantity and quality of crops. Therefore, the early detection of plant diseases can avoid any losses or damage to these crops. This paper presents an image processing and a deep learning-based automatic approach that classifies the diseases that strike the apple leaves. The proposed system has been tested using over 18,000 images from the Apple Diseases Dataset by PlantVillage, including images of healthy and affected apple leaves. We applied the VGG-16 architecture to a pre-trained unlabeled dataset of plant leave images. Then, we used some other deep learning pre-trained architectures, including Inception-V3, ResNet-50, and VGG-19, to solve the visualization-related problems in computer vision, including object classification. These networks can train the images dataset and compare the achieved results, including accuracy and error rate between those architectures. The preliminary results demonstrate the effectiveness of the proposed Inception V3 and VGG-16 approaches. The obtained results demonstrate that Inception V3 achieves an accuracy of 92.42% with an error rate of 0.3037%, while the VGG-16 network achieves an accuracy of 91.53% with an error rate of 0.4785%. The experiments show that these two deep learning networks can achieve satisfying results under various conditions, including lighting, background scene, camera resolution, size, viewpoint, and scene direction.

Phishing constitutes a significant threat in the digital world, often exploiting human vulnerabilities to illicitly obtain sensitive data such as personal credentials, financial details, and private information. Misusing this information results in substantial financial loss and personal harm to victims. This study introduces an innovative approach to mitigate the risk of phishing attacks by employing machine learning algorithms to detect phishing URLs. The proposed method applies a suite of algorithms, including j48, Naïve Bayes, JRip, and Decision Table, to a robust dataset of 11,430 URLs, each with 87 extracted features. The results reveal the considerable potential of machine learning in identifying phishing threats. Furthermore, the study explores a novel server-side analysis concept where the server scrutinizes links transmitted via emails or social communication platforms such as WhatsApp, Messenger, and Instagram. The application of phishing detection algorithms filters and prevents the delivery of phishing links, thus reducing the potential harm to users. This research is poised to significantly contribute to cybersecurity by enhancing phishing detection mechanisms' accuracy and efficiency.

A successful cloud trading system requires suitable financial incentives for all parties involved. Cloud providers in the cloud market provide computing services to clients in order to perform their tasks and earn extra money. Unfortunately, the applications in the cloud are prone to failure for several reasons. Cloud service providers are responsible for managing the availability of scheduled computing tasks in order to provide high-level quality of service for their customers. However, the cloud market is extremely heterogeneous and distributed, making resource management a challenging problem. Protecting tasks against failure is a challenging and non-trivial mission due to the dynamic, heterogeneous, and largely distributed structure of the cloud environment. The existing works in the literature focus on task failure prediction and neglect the remedial (post) actions. To address these challenges, this paper suggests a fault-tolerant resource management scheme for the cloud computing market in which the optimal amount of computing resources is extracted at each system epoch to replace failed machines. When a cloud service provider detects a malfunctioning machine, they transfer the associated work to new machinery.

Vehicular ad hoc networks have played a key role in intelligent transportation systems that considerably improve road safety and management. This new technology allows vehicles to communicate and share road information. However, malicious users may inject false emergency alerts into vehicular ad hoc networks, preventing nodes from accessing accurate road information. In order to assure the reliability and trustworthiness of information through the networks, assessing the credibility of nodes has become a critical task in vehicular ad hoc networks. A new scheme for malicious node detection is proposed in this work. Multiple factors are fed into a fuzzy logic model for evaluating the trust for each node. Vehicles are divided into clusters in our approach, and a road side unit manages each cluster. The road side unit assesses the credibility of nodes before accessing vehicular ad hoc networks. The road side unit evicts a malicious node based on trust value. Simulations are used to validate our technique. We demonstrate that our scheme can detect and evict all malicious nodes in the vehicular ad hoc network over time, lowering the ratio of malicious nodes. Furthermore, it has a positive impact on selfish node participation. The scheme increases the success rate of delivered data to the same level as the ideal cases when no selfish node is present.

Non Orthogonal Multiple Access (NOMA) successfully drew attention to the deployment of 5th Generation (5G) wireless communication systems, and it is now considered a significant technology in 5G communications. The primary enhancement in 5G is the speed, which may be 100 times faster than 4G. Due to the rising number of internal or external attacks on the Network, wireless intrusion detection systems are a vital aspect of any system connected to the Internet, and 5G will demand considerable improvements in data rate and security. In this paper, we have built a simulator for NOMA and applied a dropping attack to extract a dataset from the simulation model. The accuracy for detecting dropping attacks using the extracted data after applying ML algorithms was 95.7% for LR. Furthermore, this work suggests a methodology for wireless cyberattack detection in 5G networks based on applying several ML and DL techniques such as Decision Trees, KNN, Multi-class Decision Jungle, Multi-class Decision Forest, and Multi-class Neural Network. The proposed work is implemented and tested using a comprehensive Wi-Fi network benchmark dataset. The conducted experiments resulted in an outstanding performance with an accuracy of 99% for the KNN algorithm and 93% for DF and Neural Network.

Cloud provider can maximize his profit while guaranteeing quality of service (QoS) required by clients. In this work, we propose new scheme for maximizing cloud provider profit while guaranteeing QoS for clients. Cloud provider (CP) manages all available resources on cloud market. The key objective of our scheme is extracting the optimal charging strategy for serving clients requests to maximize CP's profit by choosing proper set of requests to be served, subject to uncertain demand of service and time-varying service cost. As variety of clients classes submit their requests and offer different prices for service, this problem deserves special treatment. To tackle the uncertain service prices, Q-Iearning is proposed to extract the optimal charging policy. Numerical results show that the proposed scheme can improve the reward of CP significantly.

Securing communication in vehicular ad hoc networks (VANETs) is hampered by numerous constraints, making it more difficult. First, traditional security schemes cannot be directly applied in VANET because they consider fixed topology. Second, VANET enables dynamic spectrum access where nodes constantly change frequencies due to their high degree of mobility, resulting in severe consequences on network performance. Third, an effective security scheme in VANET needs local and continual knowledge of nodes. Last, the presence of malicious nodes and their misbehaving activities impair the safety of the drivers since they might alter the content of the sent safety alerts. With these constraints in mind, this paper presents a unique security strategy that utilizes node behaviour during message exchange as a security metric to address these issues. Through the message alert exchange phase, node behaviour is measured through the fuzzy logic framework to generate a rank for each node called trust level (BL), which describes the node’s reliability in exchanging safety messages correctly. Moreover, all messages in VANET are encrypted using the existing cryptography techniques. The proposed scheme is developed to enhance communication security in VANET, minimize the effects of malicious nodes, and improve resource utilization in VANET. Evaluation of the proposed scheme shows that it improves the performance of VANET in terms of end-to-end delay, packet delivery ratio, and packet loss ratio. According to the results, our scheme improves throughput by up to 23% and reduces end-to-end delay by up to 60%.

Design and development of an IoT-based solar

Object detection is considered a hot research topic in applications of artificial intel-ligence and computer vision. Historically, object detection was widely used in var-ious fields like surveillance, fine-grained activities and robotics. All studies focus on improving accuracy for object detection using images, whether indoor or outdoor scenes. Therefore, this paper took a shot by improving the doable features extraction and proposing crossed sliding window approach using exiting classifiers for object de-tection. In this paper, the contribution includes two parts: First, improving local depth pattern feature along side SIFT and the second part explains a new technique presented by proposing crossed sliding window approach using two different types of images (colored and depth). Two types of features local depth patterns for detection (LDPD) and scale-invariant feature transform (SIFT) were merged as one feature vector. The RGB-D object dataset has been used and it consists of 300 different objects and in-cludes thousands of scenes. The proposed approach achieved high results comparing to other features or separated features that are used in this paper. All experiments and comparatives were applied on the same dataset for the same objective. Experimental results report a high accuracy in terms of detection rate, recall, precision and F1 scorein RGB-D scenes.