Using Machine Learning techniques to improve Intrusion Detection Accuracy

In recent years, network intrusions have emerged in an endless stream and have generated a lot of financial losses. An intrusion detection system (IDS) is used to detect network attackers. Most current intrusion detection systems are based on a specific predefined pattern (signature) that matches known attack functions. The main limitation of the signature-based approach is that it does not recognize new attacks, even small changes in known vulnerabilities. In addition, in order to keep the extensive signature database up-to-date, an important administrative expense is required. In this paper, based on machine learning techniques, k-means clustering algorithm and an alternative method of support vector machine classification algorithm, it can automatically construct the distribution of normal packet payload and detect its deviation. Our method shows that the proposed hybrid algorithm provides significant detection accuracy than the most used open-source Snort system.