Deep Learning for Web Intrusion Detection

Detecting and responding to cyber-attacks on web applications efficiently remains a critical challenge due to their vulnerability and accessibility over networks. This study explores the feasibility of employing unsupervised or semi-supervised methods, leveraging the Robust Software Modelling Tool (RSMT), an open-source tool for observing and characterizing web application behaviors. Initially, the study examines the use of RSMT to detect web attacks through encoded representations and reconstructed call graphs using a stacked denoising autoencoder. Subsequently, the effectiveness of these methods is evaluated using limited labeled data, demonstrating efficient and accurate detection capabilities, particularly when employing the Long Short-Term Memory (LSTM) algorithm. This research contributes to advancing intrusion detection systems by proposing a robust approach that enhances the security posture of web applications against cyber threats.