Methods for Improving the Quality of Classification on Imbalanced Data

The subject of the study is methods of balancing raw data. The purpose of the article is to improve the quality of intrusion detection in computer networks by using class balancing methods. Task: to investigate methods of balancing classes and to develop a classification method on imbalanced data to increase the level of network security. The methods used are: methods of artificial intelligence, machine learning. The following results were obtained: Class balancing methods based on Undersampling, Oversampling and their combinations were studied. The following methods were chosen for further research: SMOTEENN, SVMSMOTE, BorderlineSMOTE, ADASYN, SMOTE, KMeansSMOTE. The UNSW-NB 15 set was used as the source data, which contains information about the normal functioning of the network and during intrusions. A decision tree based on the CART (Classification And Regression Tree) algorithm was used as the basic classifier. According to the research results, it was found that the use of the SMOTEENN method provides an opportunity to improve the quality of detection of intrusions in the functioning of the network. Conclusions. The scientific novelty of the obtained results lies in the complex use of data balancing methods and the method of data classification based on decision trees to detect intrusions into computer networks, which made it possible to reduce the number of Type II errors.