Fast kNN Classifiers for Network Intrusion Detection System

Abstract

Objective and Background: To adapt two fast kNN classification algorithms i.e., Indexed Partial Distance Search kNearest Neighbor (IKPDS), Partial Distance Search kNearest Neighbor (KPDS) and comparing with traditional kNN classification for Network Intrusion Detection. Methods/Statistical Analysis: NSL-KDD data set is used to evaluate the kNN classification, KPDS and IKPDS with 10 fold cross validation test. This experiment results shows that the IKPDS reduces the classification completion time compare with kNN and KPDS by preserving the same classification accuracy as well as the same error rate for different types of attacks. A novelistic method proposed for classifying the unknown patterns whether it is a malicious or legitimate using IKPDS algorithm. Findings: These algorithms efficiency were tested with the sample of 12597 instances and verified with actual class label. The resultsshow that 99.6% accuracy of the proposed method. Applications/Improvements: A deep analysis can be performed on DoS and Probe attacks as they are exhibiting similar characters andfeature selection techniques may also be implemented inorder to improve the accuracy and reduce the computational time.