An Efficient Machine Learning Based Classification Scheme for Detecting Distributed Command & Control Traffic of P2P Botnets

Abstract

Biggest internet security threat is the rise of Botnets having modular and flexible structures. The combined power of thousands of remotely controlled computers increases the speed and severity of attacks. In this paper, we provide a comparative analysis of machine-learning based classification of botnet command & control(C&C) traffic for proactive detection of Peer-to- Peer (P2P) botnets. We combine some of selected botnet C&C traffic flow features with that of carefully selected botnet behavioral characteristic features for better classification using machine learning algorithms. Our simulation results show that our method is very effective having very good test accuracy and very little training time. We compare the performances of Decision Tree (C4.5), Bayesian Network and Linear Support Vector Machines using performance metrics like accuracy, sensitivity, positive predictive value(PPV) and F-Measure. We also provide a comparative analysis of our predictive models using AUC (area under ROC curve). Finally, we propose a rule induction algorithm from original C4.5 algorithm of Quinlan. Our proposed algorithm produces better accuracy than the original decision tree classifier.