Applying Intrusion Detection Algorithms on the KDD-99 Dataset

Abstract

Practical task of information reliability and security is the effective intrusion detection and prevention. Open systems are vulnerable. Having in detail information about system structures, more and more sophisticated network intrusion methods could be easily developed and quickly tested. Intruders are always keeping update information about the current technology and generate new intrusion methods. There are several defense solutions against intrusions. The most common solution is Intrusion Detection System (IDS). For giving a short overview of some IDS methods, this paper applies the commonly available KDD-99 dataset for compare and discuss the IDS performance in case of different intrusion types. In this paper, the IDS performance of the J48, Random Forest, Random Tree, Decision Table, Multi-layer Perceptron (MLP) and Naive Bayes Classifier compared based on the average accuracy rate, precision, false positive and false negative performance in case of DOS, R2L, U2R, and PROBE attacks. Moreover, the focus would be on false alarm values. During the tests, the random forest algorithm produced the highest average of accuracy rate 93.77%, while the Random tree algorithm had the lowest rate 90.57%. The lowest value of false negative was produced by the decision table algorithm.