A Novel Adaptive Minority Oversampling Technique for Improved Classification in Data Imbalanced Scenarios

Imbalance in the proportion of training samples belonging to different\nclasses often poses performance degradation of conventional classifiers. This\nis primarily due to the tendency of the classifier to be biased towards the\nmajority classes in the imbalanced dataset. In this paper, we propose a novel\nthree step technique to address imbalanced data. As a first step we\nsignificantly oversample the minority class distribution by employing the\ntraditional Synthetic Minority OverSampling Technique (SMOTE) algorithm using\nthe neighborhood of the minority class samples and in the next step we\npartition the generated samples using a Gaussian-Mixture Model based clustering\nalgorithm. In the final step synthetic data samples are chosen based on the\nweight associated with the cluster, the weight itself being determined by the\ndistribution of the majority class samples. Extensive experiments on several\nstandard datasets from diverse domains shows the usefulness of the proposed\ntechnique in comparison with the original SMOTE and its state-of-the-art\nvariants algorithms.\n