Ensemble Model for Enhancing Classification Accuracy in Intrusion Detection Systems

Abstract

Currently, cases of data breaches are prevalent, partly due to the inability of the Intrusion Detection Systems (IDS) used to offer on-demand network protection through real-time intrusion detection. This study sought to develop an ideal machine learning model for enhancing accuracy in network intrusion detection to address this problem. Naïve Bayes, Artificial Neural Network, K nearest Neighbor, Support Vector Machine, and C 4.5 algorithms were trained and tested on the CIC-IDS2017 dataset using the k-folds cross-validation approach. AdaBoost, Bootstrap Aggregation, and Stacking ensemble models, using each of the five algorithms as base models, were also trained and tested on the same dataset. A comparison of the performance of the individual models and the ensemble models was done, and the best performing model was selected and tuned with respect to the number of iterations, batch size, and weight threshold to further enhance its quality and accuracy in prediction. AdaBoost ensemble model with C 4.5 as the base algorithm was found to give an effective model that could be implemented on IDS to enhance precision and recall, which translates to increased accuracy and efficiency in the classification of new instances. Depending on the volume of the data packets being transmitted on the organizational network, the model may require tuning on the batch size and the number of iterations in order to increase its accuracy, efficiency, and consistency in light of the available computational resources.