The effect of feature reduction on an ANFIS-based intrusion detection system

The key goal of an intrusion detection system is classifying the activities of a system into two main normal and intrusion activities. Various methods have been proposed to develop an intrusion detection system by researchers. The present study aimed to present an intrusion detection system based on an adaptive neuro-fuzzy inference system (ANFIS) as a strong classifier. The proposed system was assessed using standard evaluation criteria of “the percentage of false positive rate” and “percentage of true positive rate”. In the present study, we applied the NSLKDD dataset, and the percentage of false positive rate and the percentage of true positive rate for testing data was estimated at 95.65% and 0.219%, respectively. Production of an intrusion detection system by ANFIS requires a lot of processing due to lots of input, training, and testing. Therefore, attempts were made to reduce the number of inputs, for which we applied Fisher’s score method. After feature reduction, the developed intrusion detection system was trained with reduced features and eventually tested. Despite the decrease of features, the percentage of intrusion true positive rate and the percentage of false positive rates were at an acceptable level of 93.11 and 4.24, respectively.