Deep learning approaches to SQL injection detection: evaluating ANNs, CNNs, and RNNs

Abstract

In the digital era, SQL injection (SQLi) attacks on web applications pose significant threats to data integrity and security. While traditional methods such as signature-based and anomaly-based detections have some limitations, this research explores the application of neural networks in countering these attacks. Specifically, this research evaluates the performance of three primary neural network architectures: Artificial Neural Networks (ANNs), Convolutional Neural Networks (CNNs), and Recurrent Neural Networks (RNNs) for SQLi attack detection. The research methodology involves converting text-based SQL queries into numeric values suitable and compatible with the neural networks, using Term Frequency-Inverse Document Frequency (TF-IDF), tokenization, and padding. Results show that the CNN outperforms in almost all metrics, with RNNs following closely and ANNs achieving the lower results.