An Adaptive Oversampling Method for Imbalanced Datasets Based on Mean-Shift and SMOTE

Class imbalance is a challenge in different actual datasets, where the majority class contains a large number of data points, and the minority class contains a small number of data points. Class imbalance affects the learning process negatively, resulting in classification algorithms’ ignorance of the minority class. To address this issue, various researchers developed different algorithms to tackle the problem; however, the majority of these algorithms are complex and generate noise. This paper provides a simple and effective oversampling technique based on the mean-shift clustering algorithm and using the synthetic minority oversampling technique (SMOTE) of selected clusters. We conducted several experiments to compare the performance of our technique with different algorithms mentioned in the literature on three common datasets. Experimental results indicate that our technique performs better in synthesizing new samples and improves support vector machine (SVM) classification performance on imbalanced datasets.