Research on Hierarchical Clustering Undersampling and Random Forest Fusion Classification Method

Abstract

For the shortcoming of reduced generalization ability of random forests in the big data era, a classification method for hierarchical clustering of undersampled fused random forests is presented in this paper. The proposed method clusters the majority of samples through a hierarchical clustering algorithm, undersampling the samples of each cluster with a minority samples, bringing the data samples to equilibrium, and then building a random forest. This experiment used the CGSS data for 2015, compared with the classification method of random undersampled fused random forests, the prediction accuracy and F value were improved by 16% and 17%, which proved that the generalization ability of random forests was improved in this method. Based on the analysis of the method and experimental data of this paper, it is concluded that three important decision-making factors affecting commercial medical endowment insurance are family income, the using frequency of internet and age, which provide a new idea for studying the influencing factors of commercial insurance demand and predicting the commercial insurance purchase behavior.