International Journal of Digital Technology and Network Security System (e-ISSN: 3108-3307)

Autoencoders and Support Vector Machine for Zero-Day Threat Detection in Web Applications

2026-01-20T11:35:28+00:00

Zero-day attacks represent one of the most critical and complex threats to modern web-based systems, as they exploit previously unknown vulnerabilities before security patches or signatures become available. Traditional signature-based intrusion detection systems often fail to identify such attacks due to their reliance on known patterns. To address this limitation, this research proposes a hybrid intelligent detection framework that combines Autoencoders and Support Vector Machines (SVMs) for effective zero-day attack detection in web environments. The Autoencoder component operates as an unsupervised anomaly detection mechanism, learning latent representations of normal network traffic and identifying deviations that may indicate suspicious behavior. These detected anomalies are then processed by an SVM classifier, which performs supervised learning to distinguish between benign and malicious activities. The proposed framework is evaluated using widely recognized benchmark datasets, including CICIDS2017 and NSL-KDD, ensuring robustness and comparability with existing approaches. Comprehensive preprocessing techniques such as feature normalization, dimensionality reduction, and class balancing are applied to enhance model performance. Experimental results demonstrate that the hybrid Autoencoder–SVM model achieves higher detection accuracy, improved generalization, and significantly reduced false-positive rates compared to standalone machine learning and deep learning models. The findings highlight the effectiveness of integrating unsupervised and supervised learning techniques to detect evolving and previously unseen attack patterns. Overall, this study presents a scalable and resilient solution for zero-day threat detection, contributing to enhanced security of web applications and network infrastructures.

4G LTE and 5G in Military Missions: Operational Challenges and Strategic Opportunities

2026-01-30T09:49:59+00:00

Modern military operations increasingly rely on communication networks that can support high data volumes, flexibility, and interoperability across land, air, maritime, cyber, and space domains. Traditional military communication systems have delivered reliability over decades, but their voice-centric design and limited data capacity restrict their ability to support modern requirements such as real-time situational awareness, sensor fusion, and joint or coalition operations. Consequently, armed forces are progressively adopting commercial cellular technologies, particularly 4G LTE and 5G, within military communication architectures. This paper examines the role of LTE and 5G in military missions, focusing on their operational and strategic value. Comparative analysis indicates that these technologies provide advantages in deployment flexibility, cost efficiency, ease of maintenance, and interoperability with government and public safety networks. Evidence shows that 4G LTE is already a mature and operationally proven solution, widely used by the United States, NATO, and allied forces for bases, training, mission command, logistics, and civil–military support. Building on this foundation, 5G is emerging as a complementary capability, offering ultra-low latency, unified network management, and support for data-intensive multidomain operations. Together, LTE and 5G enable a phased, mission-driven modernization of military communications while preserving security, resilience, and long-term adaptability.

An Intelligent Machine Learning-based System for Detection of Phishing Domains Using AI/ML

2026-02-28T16:48:26+00:00

Phishing attacks continue to pose a serious challenge to cybersecurity by exploiting deceptive domain names that closely resemble legitimate websites. Phishing detection techniques mainly depend on static features that are chosen using traditional feature selection or ranking techniques. Existing detection techniques, including blacklist-based and rule-driven systems, are often ineffective against newly created phishing domains and zero-day attacks. In this work, an intelligent machine learning-based system is developed to identify phishing domains by analysing URL and domain-level characteristics. The proposed approach focuses on extracting lexical, structural, and statistical features that capture common patterns observed in malicious domains. Machine learning classifiers, including Random Forest and Support Vector Machine, are trained and evaluated using a publicly available dataset containing both phishing and legitimate URLs. Random Forest shows strong performance with faster execution, making it suitable for real-time applications where fast and reliable decisions are essential. Experimental evaluation shows that the Random Forest model achieves higher accuracy, precision, recall, and F1-score, and better overall performance, compared to SVM. The results indicate that domain-level feature analysis combined with machine learning provides an effective and scalable solution for real-time phishing domain detection. The findings confirm that feature-driven intelligent systems can successfully detect phishing domains that imitate the look and feel of genuine websites. The proposed system offers a scalable, accurate, and real-time solution for phishing domain detection and contributes to strengthening cybersecurity defences against evolving phishing strategies. The purpose is to give users a reliable tool that warns them instantly about suspicious websites, reduces the risk of financial and data loss, and strengthens overall trust in online activities.

The AI Arms Race in Database and System Security: Emerging Threats, Intelligent Defenses, and the Path Forward

2026-06-11T17:29:22+00:00

In the contemporary digital economy, data is the most valuable resource, and databases are primary targets of adversarial parties. However, static security solutions—whether rule-based, signature-based, or perimeter-focused—are increasingly ineffective against modern cyber threats. The rise of artificial intelligence introduced a paradigm shift in cybersecurity, turning it into a dual-purpose domain where technologies employed to build defense mechanisms are simultaneously applied to develop more sophisticated threats. This study analyzes database and system security through the prism of the AI arms race, which involves a continuous cycle of offensive and defensive escalation. It explores key security concepts, traces the evolution of cyberattacks from classic SQL injection to AI-assisted exploitation, and assess security mechanisms in relation to this development. Particular emphasis is placed on supply chain vulnerabilities, cloud environments, and compliance obligations that shape organizational security postures. To derive concrete lessons, it reviews three high-profile incidents—Equifax, Capital One, and SolarWinds. The article concludes by identifying open challenges and proposing directions for a security model capable of evolving in response to a dynamic threat environment.

Designing Explainable Machine Learning Models for Automated Essay Scoring with Interpretable Feedback in English Language Evaluation

2026-06-12T07:27:58+00:00

This work investigates the development of an explainable framework for Automated Essay Scoring (AES) that integrates advanced natural language processing (NLP) techniques with explainable artificial intelligence (XAI) to enhance transparency and pedagogical value in educational assessment. Automated Essay Scoring (AES) has emerged as a critical application of NLP in educational assessment, enabling scalable and efficient evaluation of written responses. However, most high-performing AES systems rely on complex machine learning models that operate as “black boxes,” limiting transparency and trust among educators and learners. This study proposes a framework for designing explainable machine learning models for AES that provide interpretable, fine-grained feedback in English language evaluation. By integrating XAI techniques such as SHAP, attention visualization, and linguistic feature attribution with deep learning architectures, the proposed system enhances both scoring accuracy and pedagogical usefulness. Experimental results demonstrate that the model achieves competitive scoring performance while offering meaningful explanations aligned with rubric-based evaluation. The findings highlight the importance of interpretability in improving trust, fairness, and learning outcomes in automated assessment systems.