Redundancy-driven modified Tomek-link based undersampling: A solution to class imbalance.

•An ensemble undersampling algorithm is proposed to solve class imbalance problem.•The impact of redundant noisy instances and outliers on classifiers are studied.•Baseline models, CNN and Tomek link are revised by embedding redundancy and outliers.•Elimination of negative instances is attained while preserving salient information.•Simulation of the models is performed for four classifiers and 10 real-life datasets.