Ranking of machine learning algorithms based on the performance in classifying DDoS attacks

Network Security has become one of the most important factors to consider as the Internet evolves. The most important attack which affects the availability of service is Distributed Denial of Service. The service disruption may cause substantial financial loss as well as damage to the concerned network system. The traffic patterns exhibited by the DDoS affected traffic can be effectively captured by machine learning algorithms. This paper gives an evaluation and ranking of some of the supervised machine learning algorithms with the aim of reducing type I and type II errors, increasing precision and recall while maintaining detection accuracy. The performance evaluation is done using Multi Criteria Decision Aid software called Visual PROMETHEE. This work demonstrates the effectiveness of ensemble based classifiers especially the ensemble algorithm of Adaboost with Random Forest as the base classifier. Publicly available datasets such as DARPA scenario specific dataset, CAIDA DDoS Attack 2007 and CAIDA Conficker are used to evaluate the algorithms.