A Stacked Ensemble Intrusion Detection Approach for the Protection of Information System
Olasehinde Olayemi · International Journal for Information Security Research · 2020
Cyber attackers daily works round the clock to compromise the availability, confidentiality and integrity of information system, protection of information system has been a great challenge to network administrators.Intrusion detection system (IDS) analyse network traffics to detect and alert any attempt to compromise the computer systems and its resources, stacked ensemble build synergy among two or more IDS models to improved intrusion detection accuracy.This research focus on the application of stacked ensemble to the development of enhanced Intrusion Detection Systems (IDS) for protection of information system.Relevant features of the UNSW-15NB intrusion detection dataset were selected to train three base machine learning algorithms; comprising of K Nearest Neighbor, Naïve Bayes' and Decision Tree, to build the base-predictive models. Decision Tree model with features selected by Information Gain features selection technique, recorded the highest classification accuracy on evaluation with the test dataset. Three meta algorithms; Multi Response Linear Regression (MLR),Meta Decision Tree (MDT), and Multiple ModelTrees (MMT) were trained with the predictions of base predictive models to build the stacked ensemble.Python programming language was used for the implementation of the ensemble models.The stacked ensemble recorded, improved classification accuracy of 3.0% over the highest accuracy recorded by the base models and 5.11% above the least accuracy recorded by the base models.False alarm improvement of 0.89% and 3.29% were recorded by the stacked ensemble over highest and least false alarm recorded by the base models respectively.The evaluation of this work shows a great improvement over reviewed works in literature