Performance improvement and comparison of NIDS by using RFE with C5.0 and RF Classifier

Abstract

Due to wide range of internet, attacks on network are increasing rapidly and network’s security is becoming a major concern. Intrusion detection system (IDS) monitors a system and a network for unauthorized activities. Huge amount of data can be handled by IDS in which all features may not improve the performance and results in increasing processing time. So, feature selection technique is used to remove irrelevant, redundant data. In this work, UNSW-NB15 training dataset is used, which consists of 45 instances from this by using RFE as feature search strategy and C5.0 DT as estimator 35 instances has been selected and given to RF classifier for multi-class classification. We also used RandomizedSearchCV to find best parameters inorder to improve the model performance. By using this, the model is improved upto 0.57%. Finally, the model is compared with SVM, LR, K-NN and NB. Among all, our model obtained highest accuracy of 83.52%.