Deep Learning Based Hybrid Architecture Combining BERT and TCNN for SQL Injection Threat Mitigation

SQL Injection (SQLi) is a kind of cyber-attack in which malicious SQL statements are introduced into a query so that an attacker can manipulate the database in unintended ways. It mostly takes place when user inputs are not properly validated or sanitized, and those inputs are used directly within SQL queries. These malicious queries allow attackers to execute unwanted, or malicious activities such as retrieving or altering sensitive data from a database or even gaining all levels of administrative access for accessing the entire database system. SQLIAs threats are increasing that needs proper prevention of such attacks for maintaining data integrity and confidentiality in web applications. In this proposed work, a hybrid detection method is proposed combining Bidirectional Encoder Representations from Transformers (BERT) and Temporal Convolutional Neural Networks (TCNN). The model exploits the contextual understanding capability of BERT and TCNN for feature extraction to enhance detection accuracy. The proposed work achieved accuracy of 97.8 % and has a high precision and recall value, which highlights the capability of the model in detecting and preventing SQL injection attacks. It therefore proposed work acts as a reliable solution for database security in web applications.