Accuracy of Machine Learning Algorithms in Detecting DoS Attacks Types

Abstract

Attaining high prediction accuracy in detecting anomalies in network traffic is a major goal in designing machine learning algorithms and in building Intrusion Detection Systems. One of the major network attack classes is Denial of Service (DoS) attack class that contains various types of attacks such as Smurf, Teardrop, Land, Back and Neptune. This paper examines the detection accuracy of a set of selected machine learning algorithms in detecting different DoS attack class types. The algorithms are belonging to different supervised techniques, namely, PART, BayesNet, IBK, Logistic, J48, Random Committee and InputMapped. The experimental work is carried out using NSL-KDD dataset and WEKA as a data mining tool. The results show that the best algorithm in detecting the Smurf attack is the Random Committee with an accuracy of 98.6161%, and the best algorithm in detecting the Neptune attack is the PART algorithm with an accuracy of 98.5539%, and on the average PART algorithm is the best algorithm in detecting DoS attacks while InputMapped algorithm is the worst.