Urban intersections pose significant challenges for reliable information dissemination among connected and autonomous vehicles (CAVs) due to dynamic topologies, occlusion effects, and fluctuating vehicle densities. Existing models, such as epidemic or graph-based approaches, fail to capture phase transition behavior critical for scalable communication. This study adapts percolation theory to model CAV networks, aiming to identify critical density thresholds (ρc) and optimize communication parameters for robust connectivity. A hybrid simulation framework integrates traffic flow dynamics (SUMO) and network modeling (NS-3), incorporating occlusion effects and dynamic clustering to derive ρc. Results reveal a phase transition at ρc ≈ 0.4 veh/m2 under moderate occlusion, with latency reductions of 60% (450 ms to 180 ms) above this threshold. Optimization identifies Pareto-optimal transmission range (r = 200 m) and packet rate (λ =5 Hz), achieving 92% packet delivery at 210 ms latency. The percolation model outperforms epidemic approaches, requiring 40% lower density for 95% coverage. This work establishes percolation theory as a physics-inspired tool for CAV network design, offering actionable insights for infrastructure planning and parameter tuning. Future research should integrate vehicle-to-infrastructure (V2I) elements and validate findings in real-world deployments.
Ontologies offer a powerful means of structuring medical knowledge for AI-driven diagnosis. This paper investigates the conceptual design of a framework that integrates ontologies and logical reasoning to automate the diagnosis of nutrition-related medical conditions. There is a significant gap in the literature on direct diagnosis using ontologies and logical reasoning for such conditions. By proposing a theoretical framework, this work aims to enhance diagnostic accuracy, efficiency, and reliability while providing explainable diagnoses. It is important to note that this study is part of an ongoing Ph.D. thesis and that the implementation of the proposed framework is planned in future work. This approach has the potential to improve patient outcomes and reduce healthcare costs significantly.
this paper demonstrates a brute-force attack on password based ModbusTCP implementations in PLCs and HMIs using a swarm of 20 microcontrollers. To counter this attack, a misleading LADDER-based algorithm that misleads attackers with false register values and randomized passwords is proposed. To evaluate the attack and the proposed protocol, two attack scenarios and two password lengths have been leveraged on a PLC in a LAN environment. Results show that the misleading protocol causes each attacker to receive different passwords and data, disrupting brute-force attempts while enabling attacker identification. When tested with one thread and 20 Arduino kits, each kit found a different password and received different data when reading PLC registers. This approach not only prevents unauthorized access but also allows administrators to track attackers' IP addresses, highlighting the effectiveness of deception-based defenses in industrial security.
The Internet of Things (IoT) has found applications across a wide range of sectors, including smart cities, agriculture, waste management, weather monitoring, and energy grids. Despite its growing impact, there is still room for enhancement—particularly in terms of energy efficiency. This study focuses on extending the lifespan of IoT networks by minimizing the energy consumption of sensor nodes. To achieve this, it proposes an optimal selection of Cluster Heads (CHs) that can effectively balance energy usage across the network. A hybrid meta-heuristic algorithm is introduced, combining the strengths of Simulated Annealing (SA) and the Whale Optimization Algorithm (WOA). Key performance indicators such as node density, energy distribution, ambient temperature, remaining battery power, and a defined cost function are employed to identify the most efficient CHs. Experimental results show that the proposed hybrid method outperforms existing algorithms, including WOA, Adaptive Gravitational Search Algorithm (AGSA), Genetic Algorithm (GA), and Artificial Bee Colony (ABC).
Software-Defined Networking (SDN) offers significant advantages in network management and scalability, but is vulnerable to security threats, particularly distributed denial-of-service (DDoS) attacks. These attacks can manipulate SDN flow tables and disrupt the network operations. This paper presents a comprehensive survey of supervised, unsupervised, and ensemble machine-learning techniques for DDoS detection in SDN. To the best of our knowledge, no previous study has provided such a detailed analysis within a single work. This study evaluates the effectiveness of various machine-learning models, focusing on their applications, benefits, and limitations in securing SDN environments. It reviews state-of-the-art methods and discusses performance metrics such as accuracy and computational complexity. The paper also identifies key research challenges and future directions for DDoS detection in SDN. The structure includes a review of related studies, a taxonomy of machine learning techniques, and a conclusion summarizing the findings and implications.
The increasing adoption of vehicle-to-vehicle (V2V) communication in intelligent transportation systems (ITS) brings both benefits and challenges, particularly in the areas of security and privacy. In high-traffic scenarios, traditional authentication methods such as the Elliptic Curve Digital Signature Algorithm (ECDSA) often struggle with scalability and computational demands. To address these limitations, we propose a lightweight, privacy-aware authentication scheme that integrates elliptic curve cryptography (ECC) with Schnorr signatures. This design supports pseudonym-based anonymity and forward secrecy, while also enabling efficient batch verification. Compared to ECDSA, our solution achieves a 40% reduction in computational overhead. Simulated evaluations show that it can handle up to 950 messages per second at roadside units (RSUs), lowers the success rate of replay attacks to just 0.02%, and offers improved resilience against Sybil attacks. These results highlight the proposed scheme’s strong potential as a viable authentication solution for future vehicular networks. guarantees safe, low-latency V2V authentication.
COVID-19 profoundly disrupted the global economy, underscoring the need for sustainable and resilient energy systems. In addition to traditional time-series approaches, incorporating process analytics provides a deeper understanding of energy consumption patterns during a disruptive event. This paper analyzes the behavioral interrelationships of various natural gas demand markets— residential, commercial, industrial, electric power, and transportation—across short-, medium-, and long-term planning horizons. Using monthly gas consumption data from January 2001 to May 2024, we apply Complete Ensemble Empirical Mode Decomposition (CEEMD) to extract intrinsic demand cycles and explore their co-movements. To enhance interpretability, we introduce a process mining perspective that treats demand data as event logs, uncovering structural shifts in market behavior due to COVID-19. Our findings reveal significant post-pandemic changes in inter-market dynamics, particularly in the strategic horizon, and highlight the value of integrating event-driven process insights with empirical forecasting. The results provide actionable guidance for sustainable production and integrated energy planning in the face of such future disruptions.
Lung cancer (LC) ranks among the most frequently diagnosed cancers and is one of the most common causes of death for men and women worldwide. Computed Tomography (CT) images are the most preferred diagnosis method because of their low cost and their faster processing times. Many researchers have proposed various ways of identifying lung cancer using CT images. However, such techniques suffer from significant false positives, leading to low accuracy. The fundamental reason results from employing a small and imbalanced dataset. This paper introduces an innovative approach for LC detection and classification from CT images based on the DenseNet201 model. Our approach comprises several advanced methods such as Focal Loss, data augmentation, and regularization to overcome the imbalanced data issue and overfitting challenge. The findings show the appropriateness of the proposal, attaining a promising performance of 98.95
Video dubbing is essential for breaking language barriers, but traditional methods are costly, time-consuming, and often suffer from poor synchronization. AI-based dubbing systems using Automatic Speech Recognition (ASR), Neural Machine Translation (NMT), and Text-To-Speech (TTS) models have improved efficiency; yet, ensuring speech duration alignment remains a challenge, especially in multilingual settings. This work introduces an Artificial Intelligence-based approach to enhance dubbing synchronization by leveraging silent moments, deploying multilingual TTS models, predicting speech duration, utilizing speaker-gender detection, and integrating Large Language Models for text summarization. Our method significantly improves synchronization accuracy and audio-visual coherence, resulting in a more natural and immersive dubbed experience. The modified Gender Identification model is able to achieve 93.92% accuracy. Also, the standard deviation of the audio speed-up variation in the target videos has been reduced significantly from 0.3 to 0.05.
Enhancing the Classification of Sentiments in Arabic Tweets through Word Embedding Models Sentiment analysis in Arabic tweets is applied in this research using word embedding techniques. The primary objective is to determine how to classify Arabic tweets as positive or negative. The analysis was based on a dataset of 4,546 Arabic tweets obtained from Kaggle. The word embedding model (e.g. Word2Vec and GloVe) was used to facilitate the analysis because its semantic meaning and relations of words were better than what was provided through traditional methods like TF-IDF. To analyze the data, some of the classification techniques that were used include logistic regression, decision trees, and support vector machines. The results of the study showed that SVM has the best accuracy at 87%, followed by decision trees at 80%, and then logistic regression at 85%. This demonstrates that word embedding simulation models are effective for Arabic sentiment analysis on social networking sites. With 80% decision trees, word embedding greatly aided pyramid computing’s divergence modulation, which allows for sentiment and mining analyses of Middle Eastern users’ social activities. SVM performed well at the more sensitive and sophisticated tasks of opinion and market research analysis involving the Arabic language. Even so, a more precise classification can be performed with the application of deep learning algorithms or a hybrid model in future work.
The rise of telemedicine platforms like Altibbi has transformed healthcare access, enabling patients to consult specialists remotely. One vital component of these platforms is medical question classification, which ensures that patient inquiries are efficiently directed to the appropriate specialists. However, these datasets often suffer from class imbalance, leading to biased model performance. This study explores the use of AraBERTv0.2-Twitter embeddings combined with ensemble learning techniques, Bagging and Boosting applied to various classifiers, including Logistic Regression (LR), Random Forests (RF), Decision Trees (DT), and K-Nearest Neighbors (KNN). We evaluate their performance on a dataset of Arabic medical questions from Altibbi under four experimental conditions: (1) Bagging without Synthetic Minority Over-sampling Technique (SMOTE), (2) Bagging with SMOTE, (3) Adaptive Boosting (AdaBoost) without SMOTE, and (4) AdaBoost with SMOTE. This allows us to analyze the impact of synthetic oversampling on classification performance and identify the most effective approach for addressing data imbalance. Experimental results show that Bagging with LR, without SMOTE, achieves the highest Macro F1-score (0.8468) and G-Mean (0.9077). AdaBoost with LR benefits significantly from oversampling, reaching a G-Mean of (0.8921) with SMOTE. Additionally, RF improves G-Mean from (0.8831) to (0.8893) when combined with Bagging and SMOTE. These findings highlight the vital role of ensemble methods in enhancing classification fairness and accuracy. By integrating such strategies, telemedicine platforms can significantly improve the automated classification of medical questions, ensuring patients receive timely and accurate responses from the right specialists.
Arabic spell checking faces inherent linguistic challenges due to its complex word structures and ambiguous spellings, where roots generate diverse forms and identical spellings mask multiple meanings. Current systems often prioritize technical accuracy over contextual understanding, struggling to preserve regional dialects or resolve semantic ambiguities. This paper introduces SAHDA, a hybrid framework that bridges computational methods with human linguistic expertise. By combining rule-based error detection, neural semantic analysis, and collaborative human-AI refinement, SAHDA addresses both structural errors and meaning-based ambiguities in written Arabic. Key innovations include context-aware dialect preservation to maintain regional linguistic diversity and adaptive learning techniques that enable efficient customization for specialized domains. Evaluations demonstrate SAHDA’s ability to resolve ambiguous terms and dialectal variations more effectively than existing approaches while minimizing overcorrection of valid regional expressions. As an open-source tool, the framework advances inclusive language technology by harmonizing standardization with dialectal richness— a critical advancement for supporting Arabic’s diverse written traditions in the digital age.
Ensuring the structural integrity of concrete infrastructures is critical for public safety, with crack detection being a vital task. This study presents Multi-crackNet, a deep learning-based framework for multi-class crack classification in concrete structures. The framework integrates state-of-the-art convolutional neural networks (CNNs) ResNet-50, MobileNetV3-Large, and EfficientNet-B0 fine-tuned on the SDNET2018 dataset, which contains over 56,000 annotated images of bridge decks, walls, and pavements. To improve model robustness and mitigate class imbalances, a comprehensive data augmentation strategy is employed. Model performance is evaluated using accuracy, precision, recall, F1-score, and AUC. Experimental results demonstrate that Multi-crackNet achieves a test precision of 95%, with ResNet-50 outperforming other models in classification accuracy. MobileNetV3-Large excels in computational efficiency, making it ideal for real-time applications, while EfficientNet-B0 offers a balanced trade-off between accuracy and computational cost. These results highlight the potential of Multi-crackNet as an effective and scalable tool for automated structural health monitoring, enhancing infrastructure assessment and maintenance.
With the increasing use of drones for commercial, industrial, and military applications, global positioning system (GPS)-based navigation is the most critical tool for positioning accuracy and autonomous decision-making. GPS spoofing attacks, however, are a realistic threat with the potential to mislead drones in estimating incorrect positions, thus inducing navigation failures, illegal landings, and security intrusions. This study experimented with the vulnerability of drones to GPS spoofing attacks and examined various mitigation and detection strategies. A laboratory-controlled setup with hardware equipment, such as GPS receivers, software radio (SDRs), and machine learning techniques, was used for anomaly detection. The Drone Flight Path Controller interface was used for the real-time observation and evasion of GPS spoofing attacks. The resultant system accurately identified spoofed signals with 97% detection performance and initiated a return-to-home (RTH) mode with an average of 2.3 seconds of detection latency. The outcomes emphasize the need to integrate high-level security solutions to protect drones against GPS-based cyberattacks, along with enabling safe and reliable autonomous operation.
Accurate breast cancer recurrence prediction improves patient outcomes and guides treatment strategies. This study comprehensively evaluates various machine learning models to predict breast cancer recurrence using the King Hussein Cancer Center (KHCC) clinical data. The dataset comprises 9,724 patient records with 21 clinical attributes, presenting a significant class imbalance. To address this issue, the Synthetic Minority Oversampling Technique (SMOTE) is employed to enhance model performance on the minority class. The study compares the performance of traditional classifiers—including Logistic Regression, Decision Trees, K-Nearest Neighbors, Naïve Bayes, and Multi-Layer Perceptron—against ensemble learning methods such as Random Forest, Gradient Boosting, AdaBoost, and Bagging. Model evaluation uses key metrics, including accuracy, sensitivity, specificity, and G-mean. The results demonstrate that ensemble learning models consistently outperform individual classifiers, with Gradient Boosting achieving the best balance between sensitivity and specificity. These findings highlight the effectiveness of ensemble methods in predictive tasks involving imbalanced medical datasets, contributing to the advancement of decision-support systems for breast cancer recurrence prediction.
A severe lung infection, pneumonia is a major global health concern. A precise diagnosis made early on is crucial for successful treatment. Although human interpretation of chest X-rays is labor-intensive and prone to errors, it is a common diagnostic tool for pneumonia. One interesting method is to automate this procedure using Convolutional Neural Networks (CNNs). This study demonstrates the efficiency of deep networks by building a new CNN model. A distributed training approach using TensorFlow's TPU (Tensor Processing Unit) maximizes the model training process. The suggested CNN model is parallelized with TPU's numerous cloud-based instances, which allows for effective compute parallelization and drastically cuts down on model training durations. According to the experimental results, training the CNN model with the TPU significantly reduces model training times. The architecture of this CNN constitutes various layers, including Conv2D layers, which apply convolution operations, and conv_blocks, which typically consist of several Conv2D layers with activation functions like ReLU, often accompanied by batch normalization. These components work together to process data to learn complex patterns. As a result, the CNN model's accuracy increased to 99.13%.
This study examines the integration of artificial intelligence (AI) and process mining in employee performance management at Prince Muhammad Bin Fahad University. A survey of 44 staff members assessed AI familiarity, productivity impact, and ethical concerns. Results showed 68% of postgraduate respondents and 75% of participants under age 35 highly accepted AI tools. Correlation and sentiment analyses, supported by predictive modeling, identified key factors influencing AI adoption. Strong associations were observed between AI learning tools and career alignment (r = 0.88), and between fairness and work quality (r = 0.82). Younger participants expressed optimism, while 61% of older respondents raised concerns about data privacy and bias. These findings underscore the importance of transparent, ethical, and user-centric AI systems. The research demonstrates that AI, integrated with process mining, offers strategic value for data-driven decision-making in human resource management.
Extracting relevant information from documents and websites is a time-consuming task that becomes increasingly complex as document length grows and websites expand. Manually retrieving precise information to meet user requirements from large documents and extensive websites is challenging and inefficient. In this context, Artificial Intelligence (AI) has emerged as a transformative technology, offering unparalleled capabilities for automating systems. AI has been widely utilized to streamline various processes, demonstrating its potential to simplify complex tasks, enhance accuracy, and reduce human effort. This research paper introduces "AzureIQ-RAG" Azure powered Intelligent Query and Hybrid Retrieval Augmented Generation (RAG) system with two primary features: (1) extracting and retrieving information from diverse document formats, including PDF, TXT, DOCX, etc. and enabling question-answering based on the provided document(s), (2) extracting content from website and enabling question-answering regarding the content of the website. The system automates the extraction process from both documents and websites. The extracted content is segmented, embedded using Azure OpenAI Embeddings, and indexed in Azure Cognitive Search for efficient retrieval and then a reranking mechanism is applied to improve precision and accuracy further. Then the final response is generated by passing the relevant content and user query to the Large Language Model (LLM) GPT -4 omni. Additionally, the system maintains separate session memory for both document and website searches, ensuring context retention for improved query handling and user experience. The document processing feature of the system has been tested on a dataset of over 950+ documents of various formats and achieved an accuracy of 94.77 %, and the website feature of the system is tested on 50+ different websites, achieving an accuracy of 75.22%.
The overarching objective of this study is to determine whether subtelomeric DNA methylation states directly govern the choice of telomere length maintenance mechanisms in cancer cells that depend on either alternative lengthening of telomeres (ALT) or telomerase (TERT). Prior evidence suggests that ALT+ cells frequently exhibit hypomethylated subtelomeric regions, facilitating homologous recombination, whereas TERT+ cells maintain higher levels of subtelomeric methylation that support telomerase-based elongation. By integrating BioNano DLS optical mapping data, quantum-derived probabilities for coverage and position, and transcriptomic profiling of ATRX, DAXX, and TERT, this study aims to clarify how methylation-mediated chromatin accessibility affects telomere biology in cancer.
This article investigates Sehgal Guseman contraction and develops innovative fixed-point results in the context of extended b-metric spaces. The discoveries improve and broaden previous findings in fixed-point theory, thereby contributing to the advancement of metric space research. We also provide an illustrated examples to back up our theoretical findings.
