Evaluation Of Selected Meta Learning Algorithms For The Prediction Improvement Of Network Intrusion Detection System
Olasehinde Olayemi, Olanrewaju Victor Johnson, Olufunke Catherine Olayemi
Abstract
Information security is a critical issue for many organizations, Intrusion detection system (IDS) promises potential solution to analyzing network traffics to detect and alert any attempt to compromise computer systems and its resources. Current penetration methods put previous works on IDS to a challenge. Stacked ensemble improves IDS prediction accuracy. This research focus on the Evaluation of selected Meta Learner algorithms for IDS improvement. Base-level IDS models of K Nearest Neighbour, Naive Bayes' and Decision Tree predictions were used to trained three selected meta learners' algorithms;(Meta Decision Tree (MDT), Multi Response Linear Regression (MLR) and Multiple Model Trees (MMT)). The evaluations of base-level and meta stacked models using UNSWNB15 test data show that, MDT models recorded the highest intrusion detection accuracy, closely followed by MMT meta learner stacked ensemble models, MLR recorded the least classification accuracy. MDT stacked ensemble models recorded the least misclassification rate, followed by MMT stacked ensemble and MLR recorded the highest misclassification rate. All the three Meta learners' models recorded better intrusion detection accuracy than the best accuracy recorded by each of the base-level model.