Enhancing Cybersecurity with Sequential Neural Models for SQL Query Anomaly Detection

Protecting sensitive data in cloud applications is important for users to maintain their privacy and for business operations to be operated without hassle. Structured Query Language Injection Attack SQLIA exploits web vulnerabilities, allowing unauthorized access to the databases and thus disrupting the applications' operations. Traditional defensive measures such as firewalls and signature-based detection prove ineffective against evolving SQLIA attacks. The proposed hybrid algorithms of Memory Augmented Temporal Long Short-Term Memory MATLSTM achieve an accuracy of 95.2%. Using memory-augmented techniques in sequential deep models such as LSTM, MAT-LSTM, and GRU further enhances the adaptability toward the dynamic pattern of attacks. The advanced techniques for feature extraction are Term Frequency-Inverse Document Frequency (TF-IDF), entropy analysis, and Principal Component Analysis (PCA), which boost the detection by compressing the dimension of the query.