Evolutionary Data Preprocessing to Alleviate Class Imbalance

Intrusion detection technology for network attacks is developing rapidly with the development of artificial intelligence technology. Recently, machine learning-based methods that can detect new types have been developed. To improve classification performance rare classes in intrusion dataset, we study efficient data preprocessing method based on learning. The UNSW-NB15, a well-known used experiments. dataset includes 9 attack and has severe class imbalance overlap, so it difficult to above certain level. by adjusting number instances needed. SMOTE techniques genetic algorithms are optimize ratio between training dataset. computation time reduced creating samples only few percent UNSW-NB15 Many datasets generated small according randomly ratios. experiments conducted these datasets. A combining results experiments, regression model best tuple ratios searched applying as fitness function algorithm. D-S-1G combination exhibited among test results. It consists decision tree classifier support vector regressor (SVR). As result, was significantly reduced, optimal showed better than experimental original found result each relies heavily type classifier.