Minority Oversampling Technique for Imbalanced Data
Date Shital Maruti · 2015
Abstract — Today, solving imbalanced problems is difficult task as it contains an unequal distribution of data samples among different classes and poses a challenge to any classifier becoming hard to learn the minority class samples. Unequal distribution of data samples among many classes confuses supervised learning based classifier as it makes challenging to learn minority class samples.Generating synthetic minority class samples tries to balance the sample distribution between minority and majority classes. To handle imbalanced learning problem, proposed work finds minority samples which are difficult to learn and computes Euclidean distance between nearest majority class samples. Using clustering approach and weighted minority class samples it generates synthetic samples for oversampling purpose. Proposed work will evaluate this approach on real & artificial datasets.