Detecting SQL Injections using Deep Learning

Abstract

SQL injection attacks remain one of the biggest threats to web applications, because they allow the attacker to gain trusted access to data without authorization, which can lead to irreparable damages. As part of this project, we examined how deep learning and machine learning can aid in detecting these attacks automatically. In total, we built and evaluated six models: Logistic Regression, Support Vector Machine (SVM), Multilayer Perceptron (MLP), Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), and BERT (Bidirectional Encoder Representations from Transformers). Overall, BERT achieved the highest scores in accuracy, precision, recall, and F1-score. This demonstrates that transformer-based models such as BERT have a better understanding of SQL query structures, which makes them efficient in detecting complex attacks. This study shows how deep learning, especially BERT, can improve web application security.