An Efficient DDoS Attack Detection in SDN using Multi-Feature Selection and Ensemble Learning

Abstract

Software-defined networking (SDN) has significantly enhanced network agility by decoupling network control from hardware devices. This architectural shift exposes SDN infrastructures to increased risks from Distributed Denial of Service (DDoS) attacks, which can severely disrupt network services and render them unavailable to legitimate users. Despite the existence of various techniques for detecting specific types of DDoS attacks, there remains a need for more comprehensive work capable of accurately identifying multiple attack categories. We have addressed the gap by proposing a robust DDoS attack detection method in the SDN. We utilize advanced machine learning (ML) algorithms to analyze the InSDN dataset, employing advanced feature extraction techniques to improve detection accuracy and efficiency. Three feature extraction methods, Se-lectKBest, ANOVA F-value scores, and feature importance scores from RandomForest Classifier, were used to select the best ten features from 81. This feature selection, combined with ensemble learning, achieved an accuracy of 99.9%. The Receiver Operating Characteristic (ROC) curve, confusion matrix, and K-fold cross-validation were used to evaluate the performance of the proposed model.