Feature Optimization and Performance Improvement of a Multiclass Intrusion Detection System using PCA and ANN

Abstract

There are several bottle necks in the process of high speed intrusion detection, of which large dimensionality is one of the major problem. We have employed the Principal Component Analysis (PCA) algorithm to handle this problem, through which we have improved the performance of the Artificial Neural Network (ANN) classifier for intrusion detection. With the help of PCA we were able to identify the top 15 out of 41 features among the feature set of KDD cup 1999 data set, and noticed an improvement of over 62% in the training time of ANN. The Multi Layer Perceptron Neural Network improved the accuracy even after the feature reduction.