A Comparison Study between Different Sampling Strategies for Intrusion Detection System of Active Learning Model

Abstract

Active learning aims to train an accurate model with minimum cost by labeling the most informative instances without compromising the model performance. So, choosing an efficient criterion for instance selection is the most important step. Sampling stage is the main issue in active learning for many problems such as intrusion detection system. There are many methods for sampling stage to select the informative instances, but what the method should be used to provide the most accurate to the Intrusion Detection System (IDS). So, we made a comparison between three of these methods, uncertainty sampling, Query By Committee (QBC) and expected model change. The contribution of this study is analyzing and examining three of common strategies that used to select the most informative instances to determine the best one of them. The experimental result showed that the expected model change method achieved the highest accuracy compared with uncertainty sampling and query by committee methods.