Comparison and Analysis of Data Mining Techniques for Intrusion Detection

Abstract

This research investigates the use of data mining techniques for intrusion detection. Decision Trees, Artificial Neural Network, Naïve Bayes and Support Vector Machine are the strategies examined. The dataset used was collected from different online repositories where they are made available as open-source data. The sample size used were 910, 10077, 10679 instances of KDDCUP’99, NSL-KDD-train and CICIDS2017 respectively, selected using random sampling technique, The Weka workbench tool was used for preprocessing and creation of the intrusion detection system.  The accuracy, speed, and scalability of the techniques, among other criteria, are taken into account and contrasted. The study also examines the best method for detecting intrusions in dynamic networks and various applications.  The performance of each technique in terms of accuracy, precision, recall, and F1-score is also examined in this research. The results revealed that the Decision Tree performs better with accuracy, precision, recall and F-measure of 99% than the other classifiers Support Vector Machine, ANN, NB and KNN in most of the tests on the three different datasets. Both Decision Tree and ANN classifier showed superior performance in detecting attacks. In conclusion, this paper reveals that artificial neural networks are the most accurate data mining technique for intrusion detection. However, in terms of implementation, Decision Tree classifier take a very short time to implement compared to ANN which takes a very long time to be implemented. Finally, this research recommended that employing these optimizing techniques to develop an intrusion detection model has a better accuracy rate.