Efficient Feature Selection for Intrusion Detection Systems

Intrusion detection systems (IDSs) monitor network traffics to find suspicious activities, such as an attack or illegal activities. These systems play an important role in securing computer networks. Due to availability of irrelevant or redundant features and big dimensionality of network datasets which results to an inefficient detection process, this study, focused on identifying important attributes in order to build an effective IDS. A majority vote system, using three standard feature selection methods, Correlation-based feature selection, Information Gain, and Chi-square is proposed to select the most relevant features for IDS. The decision tree classifier is applied on reduced feature sets to build an intrusion detection system. The results show that selected reduced attributes from the novel feature selection system give a better performance for building a computationally efficient IDS system.