An Empirical Evaluation On Comparative Machine Learning Techniques For Detection of The Distributed Denial of Service (DDoS) Attacks

The advent of the Internet that aided the efficient sharing of resources. Also, it has introduced adversaries whom are today restlessly in their continued efforts at an effective, non-detectable means to invade secure systems, either for fun or personal gains. They achieve these feats via the use of malware, which is both on the rise, wreaks havoc alongside causing loads of financial losses to users. With the upsurge to counter these escapades, users and businesses today seek means to detect these evolving behavior and pattern by these adversaries. It is also to worthy of note that adversaries have also evolved, changing their own structure to make signature detection somewhat unreliable and anomaly detection tedious to network administrators. Our study investigates the detection of the distributed denial of service (DDoS) attacks using machine learning techniques. Results shows that though evolutionary models have been successfully implemented in the detection DDoS, the search for optima is an inconclusive and continuous task. That no one method yields a better optima than hybrids. That with hybrids, users must adequately resolve the issues of data conflicts arising from the dataset to be used, conflict from the adapted statistical methods arising from data encoding, and conflicts in parameter selection to avoid model overtraining, over-fitting and over-parameterization.