A multi stage approach to handle class imbalance:An ensemble method
Shiva Prasad Koyyada, Thipendra P. Singh · Procedia Computer Science · 2023
An imbalanced learning problem is a challenge that has grabbed the interest of academics and industry. Class imbalance is a widespread problem in machine learning due to the unavailability of data in a specific category. Several under-sampling and oversampling methods handle class imbalance; however, finding the correct pattern heavily depends on the sub-sampling. This paper proposes a two-stage approach with a data sampling method where Mutually Disjoint Data Sets(MDS) created from majority samples and binded with minority samples. A set of heterogeneous machine learning algorithms were applied on these binded sub-samples to create ensemble models. MDS ensembles give importance to the minority samples with a double voting method. The proposed method has an improvement of 3.78, 13.78 percent of average recall on the train and test data compared to best-base model results