| Sumario: | SQL Injection (SQLi) remains one of the most critical security threats to modern web applications, exploiting vulnerabilities in database query handling to gain unauthorized access to sensitive data. Traditional signature- based detection methods often fail to identify novel or obfuscated attack patterns. This paper proposes a multi- stage hybrid machine learning framework that integrates three key components: (1) data augmentation using a Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP) to generate realistic malicious queries and balance class distributions; (2) contextual feature extraction using a Bidirectional Encoder Representations from Transformers (BERT) model to capture the semantic and syntactic relationships within SQL queries; and (3) classification through a hybrid Long Short-Term Memory (LSTM) and Random Forest (RF) model that refines sequential dependencies and enhances classification accuracy. Experimental evaluations demonstrate that the proposed framework achieves an accuracy of 95.1% and an area under the curve (AUC) of 0.988, outperforming conventional detection methods. The results confirm the effectiveness of integrating generative modeling, deep contextual representation, and hybrid classification for robust SQLi detection. Future work will focus on improving the generative data augmentation component and optimizing the architecture for real-time deployment in Web Application Firewalls (WAFs). The proposed framework provides a strong foundation for advancing machine learning–based SQLi detection in practical security applications.
|
|---|