ARCHIVES

Database systems represent the backbone of modern digital infrastructure, storing sensitive organizational, financial, and personal data. The escalating sophistication of database attacks —including SQL injection, privilege escalation, insider threats, and anomalous query flooding — demands intelligent, proactive security mechanisms that transcend traditional signature-based detection. This paper presents a comprehensive framework for intelligent database attack detection leveraging machine learning (ML) techniques and query behaviour analysis. The proposed system captures and analyses SQL query patterns, user access behaviours, temporal anomalies, and structural deviations to construct adaptive detection models. Algorithms including Random Forest, Long Short-Term Memory (LSTM) networks, Isolation Forest, and ensemble methods are evaluated on benchmark datasets including the CICIDS-2017 and KDD Cup 99 database-relevant subsets. Experimental results demonstrate detection accuracy exceeding 97.3% with a false positive rate below 2.1%, outperforming conventional intrusion detection approaches by a significant margin. The framework further incorporates explainable AI (XAI) via SHAP values to enhance transparency, enabling security analysts to interpret model decisions in real-time. The paper also discusses deployment considerations in cloud-hosted and hybrid database environments, contributing both theoretical insights and a practical roadmap for next-generation database security systems.