Mass ratio variance majority cleansing and minority oversampling technique for class imbalanced

Abstract

A sampling method is one of the basic methods to deal with an imbalance problem appearing in machine learning. A dataset having an imbalance problem has a noticeably skewed distribution among different classes. There are three types of sampling techniques to solve this problem by balancing class distributions, undersampling technique, over-sampling technique, and combined sampling technique. In this research, the mass ratio variance scores of each data point of the same class are computed and used to remove noise from a majority class and synthesise instances from a minority class. The results of this proposed sampling technique improve recall over standard classifiers: a decision tree, a random forest, Linear SVM, and MLP on all synthesised datasets. Performances are reported on synthesised datasets and UCI datasets via three measures: Precision, Recall, and F1-score. Moreover, Wilcoxon signed-rank tests are used to confirm the improved performance.