Optimized Feature Selection with k-Means Clustered Triangle SVM for Intrusion Detection

With the rapid progress in the network based applications, the threat of attackers and security threats has grown exponentially. Misleading of data shows many financial losses in all kind of network based environments. Day by day new vulnerabilities are detected in networking and computer products that lead to new emerging problems. One of the new prevention techniques for network threats is Intrusion Detection System (IDS). Feature selection is the major challenging issues in IDS in order to reduce the useless and redundant features among the attributes (e. g. attributes in KDD cup'99, an Intrusion Detection Data Set). In this paper, we aim to reduce feature vector space by calculating distance relation between features with Information Measure (IM) by evaluating the relation between feature and class to enhance the feature selection. Here we incorporate the Information Measure (IM) method with k-means Cluster Triangular Area Based Support Vector Machine (CTSVM) and SVM (Support Vector Machine) classifier to detect intrusion attacks. By dealing with both continuous and discrete attributes, our proposed method extracts best features with high Detection Rate (DR) and False Positive Rate (FPR).