Building an Ensemble Learning Based Algorithm for Improving Intrusion Detection System

Abstract

Intrusion detection system (IDS) alerts the network administrators against intrusive attempts. The anomalies are detected using machine learning techniques such as supervised and unsupervised learning algorithms. Repetitive and irrelevant features in data have posed a long due to speed bump in efficient network traffic classification. This issue could be resolved by reducing the dimensionality of feature space using feature selection method wherein it identifies the important features and eliminates irrelevant ones. An intrusion detection system named Least Square Support Vector Machine (LSSVM-IDS) is built using this feature selection algorithm. It is tested on intrusion detection data set like KDD Cup 99, NSK-KDD and Kyoto 2006+ data set. This LSSVM machine has accuracy of 95%, i.e. it has predicted correct output with 95% of time. To avoid the intrusion detection system from getting obsolete, to adapt it with newer attack resistance feature and also to make it less expensive, we applied ensemble learning algorithm on UNSW-NB15 data set, using a stacking classifier method. We have combined random forest, support vector machine and Naive Bayes methods using logistic regression as meta-classifier and have achieved 95% accuracy.