SMOTified-GAN for Class Imbalanced Pattern Classification Problems

Class imbalance in a dataset is major problem for classifiers that results poor prediction with high true positive rate (TPR) but low negative (TNR) majority training dataset. Generally, the pre-processing technique of oversampling minority class(es) are used to overcome this deficiency. Our focus on using hybridization Generative Adversarial Network (GAN) and Synthetic Minority Over-Sampling Technique (SMOTE) address class imbalanced problems. We propose novel two-phase approach involving knowledge transfer has synergy SMOTE GAN. The unrealistic or overgeneralized samples transformed into realistic distribution data by GAN where there not enough available process them itself effectively. named it SMOTified-GAN as works pre-sampled produced rather than randomly generating itself. experimental prove sample quality been improved variety tested benchmark datasets. Its performance up 9\% from next best algorithm F1-score measurements. time complexity also reasonable which around $O(N^2d^2T)$ sequential algorithm.