A Proposed Approach to DDoS Attacks Detection on SDN Using Machine Learning Technique

Abstract

Software-defined network (SDN) is becoming very popular because of numerous advantages it offers, such as scalability, centralized management, better performance and other parameters. However, it suffers from many security threats including DDoS attacks. These attacks are performed more on the controllers of SDN, which is the most important component in this new network architecture. When a DDoS attack is performed in the controller it becomes over-engaged and its power to communicate with network applications and devices is reduced drastically. Due to this overloading, the switch flow tables become exhausted and the overall network performance suffers and reduces to a bare minimum, and even stops at many times. In the proposed work, an approach to DDoS attacks in a software-defined network is described. About 130,000 feature vectors will be acquired from SDN for experimentation under normal traffic and expected DDoS-based traffic. Using feature selection methods, a proposed data set will be created. The nature of the feature selection method is to shorten the training time and prediction time, simplification of models and facilitate the interpretation of results. The original feature vector set and the data set filtered by the feature selection method will be trained with classifiers, namely naïve Bayes (NB), k-nearest neighbor (k-NN), artificial neural network (ANN) and support vector machine (SVM). The test results will be compared against existing researches. The aim of the experiments will be to obtain efficient results in terms of DDoS detection in SDN and pruning of time and processing loads.