Adaptive Hybrid Sampling Algorithm Based on BIRCH Clustering
Xuanrui Xiong, Yang Huang, Yuan Zhang, Fan Zhang, Yumei Jia, Juan Xi · 2021 IEEE 5th Information Technology,Networking,Electronic and Automation Control Conference (ITNEC) · 2021
Category imbalance has always been a hot spot in data analysis and research, where large differences between data samples can limit the performance of classifiers and lead to low sample detection rates. Resampling is currently a common method to deal with imbalance problems, aiming to balance the data by increasing the minority class samples or reducing the majority class samples. A hybrid sampling algorithm based on BIRCH clustering (HSAB) is proposed in this paper. By using BIRCH clustering to improve the shortcomings of the SMOTE algorithm, adaptively adjust the weights in different sparse regions of minority classes to improve the imbalance within and between classes. Gaussian mixture model (GMM) is used to cluster majority classes to remove redundant samples and construct a balanced dataset. The experimental results show that the balanced data set obtained by the algorithm can be effectively used for classification learning, and it provides a good strategy for classification of unbalanced data.