MLP-Based Undersampling Technique for Imbalanced Learning

The imbalanced learning problem is becoming pervasive in today's data mining applications. This problem refers to the uneven distribution of instances among the classes which poses difficulty in the classification of rare instances. Several undersampling as well as oversampling methods were proposed to deal with such imbalance. Many undersampling techniques do not consider distribution of information among the classes, similarly some oversampling techniques lead to the overfitting or may cause overgeneralization problem. This paper proposes an MLP-based undersampling technique (MLPUS) which will preserve the distribution of information while doing undersampling. This reduction can be done on the basis of stochastic measure evaluation. Experiments are performed on 10 real world data sets for the evaluation of performance of MLPUS.