Stacked Autoencoder-Based Deep Unsupervised Learning Approach for SQL Injection Attack Detection

In an era marked by escalating cyber threats, the detection and prevention of SQL injection attacks remain pivotal to the security of database-driven systems. Currently, several methods have been introduced to automatically detect the SQLIA which are based on the rule-based method, machine learning, and deep learning. This article presents a pioneering approach to SQL injection attack detection by harnessing the capabilities of autoencoders, stacked autoencoders, and K-means clustering. Autoencoders serve as the foundation of our proposed model, enabling the extraction of intricate data patterns that are characteristic of SQL injection attacks. Stacked autoencoders elevate this process by extracting hierarchical and abstract features, enhancing both detection accuracy and the model's ability to adapt to emerging attack strategies. The integration of K-means clustering complements this feature extraction mechanism, allowing for the categorization of learned representations into meaningful clusters. This unsupervised clustering technique not only streamlines data exploration but also facilitates the identification of anomalous attack patterns, even in the absence of labeled attack samples. The performance of proposed model is measured for KDDCUP dataset, NSL-KDD, and CIC-CSE-IDS 2018 datasets.