K-nearest neighbor and boundary cutting algorithm for intrusion detection system

Abstract

Intrusion detection system is used for securing computer networks. Different data mining techniques are used for intrusion detection system with low accuracy and high false positive rate. Hicuts, HyperCuts, and EffiCuts are decision tree based packet classification algorithm which performs excellent search in classifier but requires high amount of memory. So in order to overcome these disadvantages, new approach is provided. In this, we present a hybrid approach for intrusion detection system. Boundary Cutting Algorithm and K-Nearest Neighbor using Manhattan and Jaccard coefficient similarity distance is used for high detection rate, low false alarm and less memory requirement. KDD Cup 99 dataset is used for evaluation of these algorithms. Result is evaluated using KDD CUP 99 dataset in term of accuracy, false alarm rate. Majority voting is done. This approach provides high accuracy and low memory requirements as compare to other algorithm.