Decision tree for credit scoring and discovery of significant features: empirical based on Chinese microfinance for farmers

Abstract

For the tens of thousands of farmers’ loan financing, it’s imperative to find which features are the key indicators affecting the credit scoring of rural households. In this paper, C5.0, CHAID and C&RT three models are used to screen the key indicators affecting farmers’ credit scoring, and 2044 farmers’ microfinance data from 28 provinces in China are applied in the empirical study. The empirical results show the classification accuracy of C5.0 is better than CHAID and C&RT in both the training set and test set, thus finally use the feature subset selected by C5.0. Six key features screened from 44 attributes by C5.0, which have significant influence on credit scoring of farmers, namely, education level, net income each year/per capita GDP, education cost of children each year, Residence type, residential year, relationship with cosigners.