Abstract

In data mining, imbalance learning is a challenging\ntask due to the intrinsic properties of the imbalance datasets. An\nimbalance data consists of unequal ratio instances in the classes.\nTo address the limitations of imbalance data, we propose a novel\nalgorithm dubbed as, In Excess Less Than (IELT) sampling\ntechnique taking into account both under sampling ad over\nsampling. In fact, our algorithm is capable of restructuring the\noriginal dataset at a very high conceptual level to alleviate the\nproblems in the class imbalance. We conduct the empirical\nbenchmark experimental setup using 15 datasets of varying class\nimbalance level. The proposed IELT approach performs\neffectively than the compared five algorithms on three evaluation\nmetrics.