THE METHODS FOR QUANTITATIVE SOLVING THE CLASS IMBALANCE PROBLEM

Context. The problem of recovery the classes' balance in imbalanced samples is solved to increase the efficiency of diagnostic and recognition models. Objective. The purpose of the work is to modify the existing method of recovery classes' balance and to conduct comparative analysis of performance indicators with some modern methods. Method. The proposed data preprocessing method is based on combining the undersampling and cluster-analysis technologies. The method has allowed restoring the balance and reducing the sample while maintaining important topological properties of the sample, high accuracy and acceptable operating time. Results. The software that implements in proposed method has been developed and used in the computational experiments on the study of method's properties and comparative analysis with other methods of restoring classes' balance. Conclusions. The experiments confirmed the efficiency of the proposed method and its implemented software. The method has allowed reducing the majority class to the size of the minority class, thus reducing the training sample (the sample is considered imbalanced if the size of the minority class is less than 10% of the original sample size), while demonstrating the best indicators of model accuracy and comparable sampling speed. It can be recommended for the practical application in solving problems of imbalance data for diagnostic and recognition models.