A novel region adaptive SMOTE algorithm for intrusion detection on imbalanced problem

Machine learning techniques play a crucial part in intrusion detection and greatly change the original intrusion detection methods. How to use machine learning technologies to achieve better detection results is important. However, due to defects in the machine learning algorithms and the data imbalance problem between the attack behaviors and the normal behaviors in the network, the detection rate of low-frequent attack behaviors cannot be effectively improved. In order to solve this issue, from the consideration of data level, a novel Region Adaptive Synthetic Minority Oversampling Technique (RA-SMOTE) is proposed. Three different types of classifiers, including support vector machines (SVM), BP neural network (BPNN), and random forests (RF), are used to test the effectiveness of the algorithm. Empirical results test on DSL-KDD dataset show that the proposed algorithm can effectively solve the class imbalance problem and improve the detection rate of low-frequent attacks.