Classifier Rank Identification using Multi-Criteria Decision Making Method for Intrusion Detection Dataset

Network intrusion detection system (NIDS) tracks network traffic for suspicious activity and policy violations. It generates alerts whenever such activity found. The objective is to detect and report anomalies. Further intrusion prevention system can take action such as blocking traffic from suspected IP addresses. Classification of network traffic as is a tedious task. Existing classifiers are suffered by generating many/false alerts. It is paramount important to select best classification approach among set of available approaches. KDD 99 is the benchmark dataset utilized to test the classification capabilities of classifiers. However, many classifiers generate similar results by measuring performance on various criteria. Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) is a traditional multi-criteria decision making (MCDM) approach which is widely used to rank classifiers from number of options that are assessed on various criteria. In this work, KDD 99 dataset is applied as input to bayes net, naive bayes, NB updateable, random forest, oneR, zeroR, adaboostM1, decision stump, J48 and decision table classifiers. The performance of each classifier is measured using 10 different criteria’s such as accuracy, misclassification, RA error, RMS error, false positive rate, f- measure, precision, RRS error, mean absolute error and recall. In order to test the effectiveness of proposed approach weka utility is utilized for classification and classifier performance result are supplied to the TOPSIS. An application is designed to implement TOPSIS method using python. It is observed that J48 secured at the top position with performance score 0.5829.