![]() | Only 14 pages are availabe for public view |
Abstract This thesis presents handles learning from imbalanced data problem. In this work, we provide a comprehensive analysis for the most well known over-sampling technique named Synthetic Minority Over-sampling Technique (SMOTE). We provide a mathematical and an experimental analysis for the distribution of generated examples using SMOTE and measure how they diverge from the original distribution. We develop several new synthetic oversampling techniques that generate synthetic examples closer to the original distribution than SMOTE generated ones and hence resulting in better classification performance |