Detection Mechanism Using Transductive Learning and Support Vectors for Software-Defined Networks

Abstract

SDN has come up as a promising technology for a future network as a logically centralized controlled framework along with its physically distributed architecture isolating the control plane from sending data moving the entire choice capacity to the regulator. SDNs are turning out to be significant because of scalability, adaptability and testing. As SDN needs overhead for operation, it makes it as a target of Distributed Denial of service (DDoS) attacks. The extensive review in the existing literature survey provides results for small footprint of dataset causing over fitting of the classifier. In the survey it is also been observed that the KNN based algorithms to detect DDOS attacks are lazy learners resulting in the noisy data. This paper proposes a Dual Probability Transductive Confidence Machines and Support Vector Machine (DPTCM-SVM) classifier to avoid the over-fitting for detecting DDoS in SDN. The results generated for detection are more than 98% for all the attack classes making it an Eager Learning System which requires less learning space unlike the Lazy Learning Systems.