Cybercrimes are consistently advancing in complexity. Cyber threat intelligence (CTI) is useful for analyzing, evaluating, and drawing conclusions for areas such as indicators of compromise (IOCs) and tactics, techniques, and procedures (TTPs). However, valuable CTI is not always utilized effectively, particularly from the unregulated Dark Web. Manually processing this vast and complex data is unrealistic. This research uses artificial intelligence (AI) and machine learning (ML) to provide automation and process a sizable amount of information in a reasonable time. Specifically, it will use the DUTA-10K Darknet dataset to develop ML models based on the Random Forest, Support Vector Machine, and XGBoost algorithms. These models distinguish illicit and benign addresses, along with their classes and language categories. The resulting models showed reliable performance, with XGBoost consistently producing more effective models overall for predicting illicit status and language.
Pathfinding algorithms are fundamental to autonomous navigation, traditionally computing optimal routes based on defined start and goal locations using heuristic cost functions. While effective, these algorithms become computationally intensive in dynamic environments due to the need for exhaustive search. Gradient-based optimization methods, such as Broyden-Fletcher-Goldfarb-Shanno (BFGS), offer a reactive alternative by making local decisions based on the current position; however, they are susceptible to local minima and increased computational burden in complex cost landscapes, particularly when second-order derivative information is required. This paper proposes an enhanced A^* algorithm through the introduction of a custom heuristic function designed to reduce computational load and facilitate reactive navigation. The proposed heuristic, along with three well-established heuristics, is evaluated against an augmented BFGS approach that integrates a repulsive force field to improve obstacle avoidance. Performance metrics include path length and path behavior near obstacles. Experimental results demonstrate that the A^* algorithm with the proposed heuristic achieves obstacle avoidance performance comparable to BFGS, outperforming traditional heuristics. These findings suggest the potential for the enhanced A^* to support reactive robotic navigation, motivating further exploration towards hybrid strategies that leverage the strengths of both global search and local optimization methods.
This paper introduces a novel memetic optimization approach that segments the population into multiple semi-independent subpopulations, referred to as countries, each evolving in parallel through concurrent local search cycles. The proposed Country-Segmented Memetic Algorithm (CSMA) leverages multithreading and SIMD capabilities to enhance computational efficiency without compromising solution quality. By decentralizing elitism and enabling controlled migration between subpopulations, the method achieves a better balance between exploration and exploitation. Experimental evaluations on standard benchmark functions demonstrate superior convergence speed and robustness compared to classical memetic algorithms. Additionally, theoretical scalability assessments using Amdahl’s Law, Gustafson’s Law, Karp-Flatt’s metric, and isoefficiency models confirm the algorithm’s effectiveness in parallel environments.
The rapid growth of e-commerce has transformed how businesses operate globally, creating new opportunities but also introducing significant security challenges. As online platforms process millions of transactions daily, they face increasingly sophisticated cyber threats that can compromise customer data and disrupt business operations. Security management becomes particularly complex in the global marketplace. Organizations must navigate multiple regulations, such as the European Union’s General Data Protection Regulation (GDPR) and California’s Consumer Privacy Act (CCPA), which set strict standards for data protection and privacy. Additionally, supply chain security presents critical challenges, as demonstrated by recent breaches where attackers exploited vulnerabilities in vendor systems to access larger networks. This research analyzes current e-commerce security challenges and potential solutions. We examine how organizations implement established frameworks like the Payment Card Industry Data Security Standard (PCI DSS), while also adopting newer technologies such as advanced encryption, blockchain, and AI-based threat detection. Our findings indicate that while technological solutions are essential, they must be complemented by strong human factors - particularly staff training and security awareness, as human error remains a significant factor in security breaches. Based on empirical evidence and industry practices, we propose practical security recommendations. These include implementing Zero Trust Architecture to strengthen access controls, following international security standards for better compliance, and developing ethical data handling practices. Our research emphasizes the importance of regular security assessments to address emerging threats. The findings provide organizations with practical approaches to protect sensitive data and maintain customer trust. By implementing these recommendations, businesses can strengthen their security measures while building a foundation for sustainable growth in digital commerce.
This study investigates the phenomenon of bug seeding in open-source software development, with a focus on how and when bugs are introduced into the source code. Previous research often assumes that the bug is introduced in the commit preceding the bug fix. However, our large-scale observational study on the OpenStack project challenges this assumption. By analyzing 459 bug reports and tracing their origins, we reveal that only 50
Text classification, a fundamental technique in information retrieval, categorizes textual data to streamline search operations, significantly enhancing efficiency. This research introduces CNN, a novel hybrid deep learning model that advances text classification by synergistically combining the strengths of transformer models, CNNs, RNNs, and LSTMs. The model stands out with its innovative design, inputting ‘input ids’ and ‘attention mask’ into a pretrained transformer, followed by a series of operations leading to robust feature extraction. CNN outperforms baseline models, demonstrating superior accuracy in text classification. Extensive experimentation on both Chinese and English datasets substantiates this, with the model achieving the highest accuracies of 94.54
Cyberattacks continue to pose a serious threat to organizational and national security, with recent years showing a dramatic rise in the number and complexity of these attacks. In 2025 alone, ransomware attacks increased by 126
An expert system is proposed to predict the risk of adverse side effects in patients taking antidepressants. The system predicts the risk of seven adverse drug reactions (serotonin syndrome, nausea, suicidal ideation, somnolence, gastrointestinal symptoms, sexual dysfunctions and hepatobiliary disorders) based on physiological information, including sex, age and pharmacological factors such as type of antidepressant and concurrent medication. The risk of side effects is categorized into four degrees: none, low, medium, and high. The effectiveness of the proposed expert system was validated using 54 patient profiles compiled from databases with high classification rates, especially for hepatobiliary disorders and gastrointestinal symptoms. The system utilizes fuzzy logic to handle the inherent imprecision to identify a patient’s age group. Furthermore, fuzzy logic does not require extensive databases and provides clinicians with an understanding of the rationale for the recommendation.
In modern computational environments, safeguarding sensitive data and preventing unauthorized access require robust and efficient authentication mechanisms. This paper presents a secure authentication protocol that addresses two key objectives: validating user legitimacy prior to network access and maintaining data integrity and confidentiality during transmission. The proposed scheme leverages lightweight cryptographic techniques including hashing, encryption, and dynamic pseudonyms to ensure mutual authentication and secure session key exchange, while preserving user anonymity and preventing identity tracking. The protocol includes a secure initialization phase, user registration, and a Mutual Authentication and Key Exchange (MAKE) process, effectively countering common threats such as replay attacks, impersonation, and man-in-the-middle attacks. Designed for low-power and resource-constrained environments, the protocol minimizes computational overhead without compromising security. Comparative analysis with existing schemes demonstrates that the proposed method achieves a strong balance between security, efficiency, and scalability, making it highly suitable for modern distributed network architectures.
As web technologies have evolved, so have the sophistication and frequency of cyber threats. Traditional cyber security measures often fail to adapt to the dynamic nature of modern web interactions and the nuanced behaviors of users. It is crucial to dynamically adjust the website s security measures based on real-time user behavior and environmental context without compromising the user experience. This research paper explores the implementation of a comprehensive user management and security system within a Django-based web application. It details the development of advanced functionalities, including user authentication, IP tracking, dynamic access control based on user behavior, and the use of middleware for security enforcement. The system’s architecture facilitates a novel approach to modifying user trust levels dynamically with direct implications on access rights, reflecting real-time security adaptations. This study aims to illustrate the practical application of combining Django’s robust framework capabilities with custom security measures to enhance application safety and user management efficiency. The results provide compelling evidence of the efficacy of dynamic security measures, highlighting several critical areas of interest in the realms of cybersecurity, user behavior analysis, and system design philosophy.
As peer-to-peer (P2P) networks grow in complexity and decentralization, enforcing secure and fair access control becomes increasingly challenging due to dynamic topologies, high churn, and limited trust anchors. Traditional token-based mechanisms, while effective in incentivizing activity, fail to capture behavioral nuances necessary for long-term governance and decision-making. In this work, we propose a novel governance token mechanism that integrates trust metrics into token distribution, enabling behavior-aware participation in blockchain governance. Based on a hierarchy of smart contracts, including the Access Control Contract (ACC), the Judging Contract (JC), and the Governance Token Contract (GTC), our system dynamically evaluates peer behavior through a cost function based on token balance, compliance rate, and peer status. Governance tokens are minted or slashed proportionally on the basis of trust scores calculated cyclically after network activity thresholds. Experimental evaluations demonstrate that our approach successfully restricts governance privileges to trustworthy peers, mitigates the influence of malicious nodes, and supports reputation recovery over time. This behavior-based model improves fairness, security, and accountability in decentralized access control.
In Vehicular Ad-hoc Networks (VANETs), participating vehicles periodically transmit basic safety messages (BSMs) containing critical information about their current status, including position, speed, acceleration and heading. BSMs are broadcast wirelessly to surrounding vehicles and help improve situational awareness. However, it also introduces critical security concerns and requires robust security measures and innovative approaches to ensure the reliability of communication within VANETs. This paper addresses the emerging threat of Traffic Congestion Sybil attacks, where malicious entities inject spurious data into BSMs to fabricate artificial traffic congestion. We implement models, using classical machine learning algorithms, to accurately detect malicious BSMs containing such spurious data. Preliminary results indicate that the proposed models can detect these sybil attacks with a high degree of accuracy and low false-positive and false-negative rates. Comparison with existing techniques demonstrate that the proposed models perform as well or better than more complex deep learning models available in the literature.
In education, the effortless analysis of test results is vital to teacher effectiveness. Instructors at all levels have relied on hardware-based OMR systems to meet this need. Although effective and requiring minimal instructor effort, this method imposes financial burdens. Costs begin immediately at the university level with the purchase and maintenance of the OMR scanner. Universities incur costs for purchasing and maintaining OMR scanners, while students must buy compatible answer sheets for each test. These expenses can be especially challenging for those with limited resources or immediate funding. This paper presents LiveTest, a highly customizable and configurable testing solution. It integrates Pictron to generate flexible answer sheets, which are graded using the dynamic OMRGrader module. Instructors can upload images of completed student answer sheets via any web browser to a predefined test for immediate grading. Results and analytics are updated in real-time, reflecting live changes as new submissions are received. The paper also details the processes of extracting, standardizing, interpreting, grading, and archiving the answer sheets.
In communities where navigating community healthcare and social services, access to clear and relevant information is more than a convenience, it is a necessity. Motivated by this need, our study explores how intelligent retrieval systems can be designed to better serve users seeking local care resources. We began with a rich directory of 1,553 community resource providers throughout Mississippi, aiming to understand how people might naturally ask for help and what ideal responses would look like. To simulate realistic interactions, we developed ten different query-reference pairs that reflect common but critical health-related concerns, ranging from mental health access to elder care services. We then implemented two information retrieval approaches: one based on Retrieval-Augmented Generation (RAG) and another using a non-RAG baseline. Using BERTScore to evaluate the semantic similarity of generated responses to our curated references, we found the RAG system to be significantly more effective, achieving an F1 score of 0.8723 compared to 0.8221 from the non-RAG model, a relative improvement of 6.1
Random or pseudo-random binary sequences are essential to various branches of science, such as cryptography, simulation, and mathematics. These values are obtained through the use of random or pseudo-random binary generators. These random strings must have a high period and high linear complexity, and must pass statistical randomness tests, to ensure their effectiveness. The development of a generator must take all of the above factors into account, and each step of the process must be meticulously controlled to ensure a satisfactory final result. Combining cryptographic components can result in an unsuitable generator. This project describes the methodology for developing a reliable binary generator. The basic components of the generator proposed in this presentation are Self-Shrinking Generators and Modified Self-Shrinking Generators, which are combined using four-variable Boolean functions.
The process of separating music and speech signals can be categorized into three scenarios. The first scenario occurs when the signals overlap only in the time domain, while the second scenario takes place when they intersect exclusively in the frequency domain. The third scenario, which is the more complex case, arises when they overlap in both time and frequency domains. To address the first scenario, a specialized frequency-domain filter is employed to distinguish the signals. When overlap occurs in the frequency domain, a time-domain filter can be utilized for segregation. However, when signals overlap in both domains, the challenge becomes more intricate, necessitating additional constraints for an effective solution. In this paper, an evolutionary voting algorithm is introduced to resolve the first two cases through enhancements of the ZCR and STE techniques, while another approach utilizing ICA is explored to tackle the third scenario with specific constraints. The paper also presents theoretical analyses and simulation results to evaluate the performance of the proposed algorithm.
This work introduces a Zero-Knowledge Proof (ZKP)-based trust verification protocol integrated within a Pyramid Tree peer-to-peer (P2P) architecture, designed to enable secure and scalable communication in decentralized systems. The protocol ensures verifiable peer authentication at each hop without revealing private information, fostering trust in dynamic, trustless environments. Our analysis shows that the ZKP message overhead grows linearly with the tree depth d and the number of puzzles k , while remaining independent of the total number of peers N . Since d ≪ n ≪ N , where n is the number of distinct resources types, the protocol ensures high security with low message complexity. This makes it ideal for large-scale, resource-constrained applications such as federated learning, IoT, and edge computing. The proposed solution offers an effective balance between trust, efficiency, and scalability, positioning it as a robust foundation for next-generation distributed architectures.
Efficient management of industrial air conditioning systems represents a critical challenge in the food industry, where environmental control directly impacts both product quality and energy consumption. This work presents an approach based on Support Vector Machines (SVM) for predicting favorable conditions for the incorporation of outdoor air, aiming to reduce the use of active climatization without compromising the required environmental parameters. More than 71,000 on-site environmental data records were used to build and validate an SVM model capable of classifying, with 99.1
Current DevSecOps automation for cloud-native applications relies on black-box machine learning models that lack transparency and formal reproducibility guarantees essential for research software validation. This paper presents the first unified framework integrating explainable artificial intelligence (XAI) throughout CI/CD pipelines while enforcing continuous reproducibility auditing. Our key contributions include: (1) novel XAI modules providing human-readable explanations for vulnerability detection, policy compliance, and anomaly identification; (2) a mathematical utility function U(Q) = α (1-S(Q)) + (1-α )R(Q) that jointly optimizes security risk and reproducibility; and (3) automated reproducibility checkpoints with environment snapshotting and result verification. Evaluation across fintech and climate simulation domains demonstrates substantial improvements: 62.5 × return on investment through proactive security prevention. The framework addresses the transparency-security-reproducibility trilemma, enabling trustworthy innovation in cloud-native research software.
The Ground Reaction Force (GRF) signals contain valuable biomechanical information that can be used for human classification tasks. This study investigates the use of Random Forest, Neural Network, and 1D Convolutional Neural Network (1D CNN) models to predict gender based on GRF data. The models are trained and evaluated on a large dataset consisting of six standardized GRF features from both feet. Experimental results demonstrate that deep learning models, particularly the 1D CNN, are highly effective in capturing complex GRF patterns and achieve superior classification performance compared to traditional machine learning approaches. These findings confirm the potential of GRF-based approaches for accurate and reliable gender classification.
