A Self-trained Support Vector Machine Approach for Intrusion Detection

Abstract

Intrusion refers to a set of attempts to compromise the confidentiality, integrity and availability (CIA) of the information system. Intrusion detection is the process of identifying such violations by analyzing the malicious attempts. Intrusion detection system is used to automate the intrusion detection process just in time or real-time and alert the system administrator for mitigating such efforts. Many researchers have been proposed several detection approaches in this context. In this paper, we adopt a semi-supervised learning-based support vector machine (SVM) approach for mitigating such malicious efforts. The proposed approach improves the learning process and the detection accuracy as compared to the standard SVM approach. Moreover, it requires less amount of labeled training data during the learning process. Our approach iteratively trains the labeled data, predicts the unlabeled data and further retrains the predicted instances. In this manner, it improves the training process and provides better result as compared to the standard SVM approach.