Multiple Model Tree Meta Algorithms Improvement of Network Intrusion Detection Predictions Accuracy

Abstract

Security of Information is a critical issue for many organizations. Intrusion Detection systems (IDSs) protect information system by analyzing network packet to determine if it is abnormal or normal. This paper applies Multiple Model Trees (MMT) stacked ensemble algorithm to improve the classification accuracy of network intrusion. The predictions of the K Nearest Neighbor, Decision Tree and Naïve Bayes intrusion detection models built with UNSW-NB15 intrusion detection training dataset served as input to Multiple Model Tree (MMT)meta learner algorithm via a tenfold cross validation to build the MMT stacked ensemble model used for the final binary classifications of the network traffics (attacks and normal) and multi-class classification into any of the nine network attacks or normal. The evaluation of all models on the testing dataset results show that MMT algorithm improves the prediction accuracy of each of the three base machine learning model predictions, It recorded the highest classification accuracy of 97.93% and lowest false alarm rate of 0.22% for the binary classification and improves the multi-class classification accuracy of all the base models prediction