Evaluating the Impact of Feature Selection Methods on the Performance of the Machine Learning Models in Detecting DDoS Attacks

Heaps of Data lie in network equipment of the organizations. To break down this information and reach some significant inferences is inconceivable for the present day IDS (Intrusion Detection System). Moreover, their signature-based defense mechanisms are ineffective in tackling emerging threats like DDoS attacks. The central premise of building Machine Learning Classifiers is to detect DDoS attacks efficiently and effectively. However, their accomplishment of machine learning models to distinguish DDoS attacks relies upon how one picks the 'relevant' and 'minimal' attributes in the network streams. The questions: "Does features choice influence the classification precision, to what extent and what is the best feature selection for boosting the performance of Machine Learning Classifiers in detecting DDoS attacks" motivate this investigation. This paper presents how to apply ML techniques in detecting DDoS attacks. There are three feature selection categories and our work demonstrates the application of the feature selection methods, one each from the three categories. In addition, five machine learning models have been utilized in our work. Random Forest classifier with Lasso-RFE, a feature selection strategy beat others. The other significance of this research paper, we believe, is a first of its kind scenario, where a real-life multidimensional dataset having diverse network traffic and recent DDoS attacks have been used unlike in most of the studies carried out to date.