An Imbalanced Data Classification Algorithm Based on Mixed Sampling

In recent years, the classification of imbalanced data sets has always been a research difficulty in the field of machine learning. Traditional classification algorithms have the problem of low recognition rate of minority samples when dealing with imbalanced data sets. To solve this problem, this paper proposes an improved random forest algorithm based on mixed sampling for imbalanced data sets. First of all, at the data preprocessing level, the improved synthetic minority oversampling technology and the Tomek-links algorithm are combined to balance the data set. Secondly, at the algorithm level, improve the traditional voting mechanism of random forests, assign different weights to each decision tree, so that decision trees with good classification effects have larger voting weights, and improve the overall classification ability of random forest algorithms for imbalanced data. Finally, the algorithm of this paper is verified on the UCI public data set. The experimental results show that the algorithm effectively improves the recognition accuracy of minority samples and improves the comprehensive classification ability of the classifier.