Enhanced Query Classification for Malicious SQL Detection Using Ensemble Learning and Bloom Filter Optimisation

SQL Injection Attacks can manipulate SQL queries to gain unauthorised access to databases, steal information, or disrupt systems. To counter such evolving attacks, the proposed work suggests an innovative detection framework that uses optimization of the Bloom filter along with a Stacking-based ensemble learning model in an innovative manner. This dual-layered approach is more scalable in real-time than existing methods, pre-filtering benign queries with minimal overhead. High-dimensional feature engineering includes lexical tokenization, bigrams, trigrams, entropy-based complexity metrics, and structural pattern heatmaps to refine query classification. Strong cryptographic SHA-256 fingerprinting strengthens the integrity of queries without putting any demands on greater processing. The proposed work achieved an accuracy and an F1-score of 95%, which is higher than CNN-BiLSTM and Adaptive Deep Forest, but at the cost of reducing computational complexity. This methodology provides a robust, scalable, and adaptive solution to the modern SQLIA detection challenges that mark a critical advancement in database security.