ACUM: An Approach to Combining Unsupervised Methods for Detecting Malicious Web Sessions

The increase in web-based attacks poses a significant risk to internet security. Detection and mitigation of malicious activity within web sessions are critical to protecting user data and maintaining the integrity of online platforms. This paper presents ACUM (Approach toCombining Unsupervised Methods), a novel approach for detecting malicious web sessions. ACUM leverages the power of unsupervised learning techniques to detect malicious and benign web sessions. By combining two unsupervised methods, including a local outlier factor algorithm and an autoencoder, ACUM effectively identifies both malicious and benign web sessions with high accuracy. The experimental results are obtained using three different datasets: a novel banking dataset, the CSIC 2010 dataset, and the WAF dataset. The experimental results of this approach demonstrate the efficacy of ACUM, outperforming existing detection methods and offering a robust solution to enhance web session security in the face of evolving threats.