A feature selection-based method for SQL injection detection using machine learning algorithms

Abstract

SQL injection (SQLi) is a serious security threat that allows attackers to access and manipulate databases through malicious input. Machine learning algorithms have shown strong potential for detecting SQL injection (SQLi) attacks. However, their performance depends heavily on the quality and relevance of the features used in training. Feature selection plays a key role in identifying the most effective, minimal set of features from the SQLi dataset. In this study, a hybrid SQLi detection method is proposed that combines feature selection with machine learning algorithms. A real-world dataset containing 13 features was first developed. Then, a hybrid Horse Herd Optimizer was developed and applied to select the most influential features before model training. Several machine learning classifiers were trained using the optimal feature set. The proposed method achieved high predictive performance, with 99.49% accuracy, 99.62% sensitivity, and 99.00% F1-score. These results were obtained using only about 45% of the original features. The reduction in feature size also improved the model's efficiency and training speed. The findings show that combining intelligent feature selection with machine learning significantly enhances SQLi detection. This approach is effective, scalable, and suitable for real-world security applications.