Research on Improving SMOTE Algorithms for Unbalanced Data Set Classification
Xiaoli Li, Qinghua Zhou · 2019 International Conference on Electronic Engineering and Informatics (EEI) · 2019
Aiming at the problem that the classification effect of unbalanced data sets is not ideal, a new method of improving SMOTE algorithm, TDSMOTE algorithm, is proposed in this paper. A small number of samples are divided into three regions, dense region, boundary region and sparse region. Different sampling methods are used for samples from different regions. Finally, random forest algorithm is used to classify the samples. Experiments were carried out on six UCI unbalanced datasets and compared with traditional algorithms. Experiments show that the TDSMOTE algorithm achieves higher G-means, F-value and AUC values, and effectively improves the classification performance of unbalanced data sets.