Fusion technique for handling imbalanced and overlapped dataset to improve the performance of existing classifiers

The effectiveness of the classifier in efficiently classifying the data is diminished by imbalanced class data. Employing a supervised method to address imbalanced data may result in the exclusion of majority classes. The overlap of class instances can complicate the learning process of the classifier. In this paper, three different algorithms were proposed to handle the imbalance problem. The proposed algorithms MCROR, RbImbD and Fused-MCRb with different working mechanisms were applied to balance the majority and the minority classes present in the given dataset. Twenty-three publicly available dataset and 12 synthesised dataset were used to evaluate the performance of the proposed algorithms. The results were calculated on different performance metrices: precision, sensitivity, F1 score and geometric mean. The values obtained for different metrices shows that the proposed algorithms significantly improved the performance of the classifier and handle the imbalance classes in the given dataset.