Wrapper-Based Feature Selection to Improve The Accuracy of Intrusion Detection System (IDS)

An Intrusion Detection System (IDS) is a system that can detect attacks on a network. IDS systems can be built using machine learning approaches. However, significant network traffic and many features cause machine learning algorithms to work slowly. Not all available features are essential for use as a reference in an intrusion detection system. Feature selection is important because it can reduce the dataset and speed up computing time in building models and detection systems. In this paper, feature selection using the wrapper method is proposed. Its performance is compared with filter-based feature selection, such as chi-square, information gain, gain ratio, and correlation-based feature selection (CFS). Based on the exploration, the results showed that the feature selection algorithm using the wrapper method was superior to other algorithms. The number of critical features discovered and the detection system's precision serve as proof of the wrapper approach. With only three features as a reference, the detection system produces an accuracy of 96.22%. It increases 5.84% from 90.38% to 96.22%. Computation time also increases because the number of features used is tiny compared to other algorithms and uses all features.