Combination of Multi Classification Algorithms for Intrusion Detection System
Maiwan Bahjat Abdulrazaq, Azar Abid Salih · 2015
Classification is one of the common tasks that are involved in data mining to build models for the prediction of future data. It performs its task by different classifier algorithms. This paper provides an approach based on information gain to determine the most distinguishing subset features of each attack class and combine multi classification algorithms which includes (Decision Tree J48, k nearest Neighbor and Naive Bays). These classifi- ers are used for the task of detecting intrusions and comparing their relative performances. The goal of this work is to analyze the performance and ac- curacy of classification algorithms in order to identify the most efficient algorithm for each attack class, and then build accurate intrusion detection sys- tem. The proposed model has been applied on KDD Cup 99 data set using 60% of them for training and 40% for testing. These experimental results show that multiple classifiers work better than a single classifier. Also, multiple classifiers are more accurate and have abilities of distinguishing among the different attacks and normal connections effectively.