Intrusion Detection Using Random Forests Classifier with SMOTE and Feature Reduction

Intrusion Detection Systems (IDS) have become crucial components in computer and network security. NSL-KDD intrusion detection dataset which is an enhanced version of KDDCUP´99 dataset was used as the experiment dataset in this paper. Because of inherent characteristics of intrusion detection, still there is huge imbalance between the classes in the NSL-KDD dataset, which makes harder to apply machine learning effectively in the area of intrusion detection. In dealing with class imbalance in this paper Synthetic Minority Over sampling Technique (SMOTE) is applied to the training dataset. A feature selection method based on Information Gain is presented and used to construct a reduced feature subset of NSL-KDD dataset. Random Forests are used as a classifier for the proposed intrusion detection framework. Empirical results show that Random Forests classifier with SMOTE and information gain based feature selection gives better performance in designing IDS that is efficient and effective for network intrusion detection.