Performance Comparison of Intrusion Detection Machine Learning Classifiers on Benchmark and New Datasets

With the tremendous growth of the Internet and the continuous increase in malicious attacks on corporate networks, Intrusion Detection Systems (IDS) have been designed and adopted by organizations to accurately detect intrusion and other malicious activities. But these IDSs still suffer from setbacks such as False Positives (FP), low detection accuracy and False Negatives (FN). To enhance the performance of IDSs, machine learning classifiers are used to aid detection accuracy and greatly reduce the false positive and false negative rate. In this research we have evaluated six classifiers such as Decision Tree (J48), Random Forest (RF), K-Nearest Neighbor (K-NN), Nave Bayes (NB), Support Vector Machine (SVM) and Artificial Neural Networks (ANN) on three different types of datasets such as NSL-KDD, UNSW-NB15 and Phishing dataset. Our results show that K-NN and J48 are the best performing classifiers when it comes to detection accuracy, testing time and false positive rate.