A two-stage classification framework for imbalanced data with overlapping labels

Abstract

Classification is one of the most significant methods in predictive analysis for categorical labeled problem. However, an accurate classification model is difficult to train for some real cases due to imbalanced samples, large fluctuating records, and overlapping class labels. For solving the above problems, in this work, we introduce a Two-Stage with Enhanced Samples (TSES) prediction framework that can balance the samples using Two-Stage classification method and increase the number of sample to make it enough for obtaining an accurate model. The proposed TSES achieves outstanding classification performance on a real case of rainfall prediction. For proving the effectiveness of TSES, we compare it with some traditional classification algorithms. The results show that it can be a promising method for the prediction problems with imbalanced data with overlapping labels.