A Machine Learning–Based Framework for Enhancing VPN Anonymity Against Traffic Analysis Attacks

Abstract

VPNs are also widely adopted to enhance privacy in the internet and secure internet users communications. In the recent past, advances in traffic analysis and machine learning algorithms have shown the vulnerabilities of the VPN systems in concealing original IP addresses despite encryption. This paper proposes a machine learning solution to reverse the approach of determining IP addresses based on traffic analysis, thereby enhancing the privacy of VPNs. A classification model was built based on a publicly accessible VPN traffic dataset in the form of a Random Forest classification model and compared to the K-Nearest Neighbors (KNN) algorithm. Statistical comparison of performance measures of accuracy, precision, recall and F1-score are conducted by independent sample t-tests. According to the experimental result, the accuracy of the proposed Random Forest model of 94.67% is much greater than that of the KNN algorithm, which is 78.58%. The findings confirm that ensemble learning algorithms are more effective in traffic analysis attack defense and increase anonymity in VPN communication.