Knowledgeable Handling of Impreciseness in Feature Subset Selection using Intuitionistic Fuzzy Mutual Information of Intrusion Detection System

One of the most promising areas of domain in research field is security because of its exponential usage in everyday commercial activities. Due to prevalence diffusion of network connectivity, there is a high demand for protection against cyber-attack which necessitates the importance of intrusion detection system as a significant tool for network security. There are many intrusion detection models available to classify the network traffic s either normal or attack type. Because of huge volume of network traffic data, these classifier techniques fail to attain high detection rate with less false alarms. To overcome the above problem, this paper introduces the potential feature subset selection model using Intuitionistic Fuzzy Mutual Information (IFMI). This model efficiently selects the optimal set of attributes without loss of information even in presence of impreciseness among attributes. This is achieved by representing each attribute in the dataset in terms of degree of membership, non-membership and hesitation. To validate the performance of the IFMI its reduced feature subset is used for classification using random forest classifier. After analyzing the feature subset, the simulation results proved that the proposed model has improved the performance of classifier for predicting the network intrusion attempts. It also helps the classification model to achieve high classification rate and reduced false alarm rate in an optimized way.