An Improved SMOTE Imbalanced Data Classification Method Based on Support Degree
Kewen Li, Wenrong Zhang, Qinghua Lu, Xianghua Fang · 2014
Imbalanced data-set Classification has become a hotspot problem in Data Mining. The essential assumption of the traditional classification algorithms is that the distribution of the classes is balanced, therefore the algorithms used in Imbalanced data-set Classification cannot achieve an ideal effect. In view of imbalance date-set classification, we propose an over sampling method based on support degree in order to guide people to select minority class samples and generate new minority class samples. In the light of support degree, it is now possible to identify minority class boundary samples, then produce a number of new samples between the boundary samples and their neighbors, finally add the synthetic samples to the original data-set to participate in training and testing. Experimental results show that the method has an obvious advantage in dealing with imbalanced data-set.