Home credit risk modelling with the fuzzy regression functions method

The assessment of credit risk has become an important issue due to the financial crises experienced in recent years. By determining the credit risk correctly, financial institutions can reduce their credit-related risks as much as possible by choosing not to give credit to individuals or institutions that apply for credit or to demand additional collateral. Credit risk assessment models are used to enable the financial institutions to make the right decision based on data. However, credit risk data generally show an unbalanced distribution with respect to the ratio of risky and non-risky individuals. In classification models, an unbalanced form results in biased predictions in favor of the majority class, and the classification accuracy remains very low for the minority class which includes risky individuals. Due to its relatively high risk and resulting serious financial losses, banks and financial institutions, in practice, attach greater importance to determining minority class cases. In this study, Fuzzy Regression Function (BRF) approach was used to evalaute the home credit risk. It is aimed to avoid the disadvantages brought by the unbalanced data set and to obtain higher accuracy estimates of the minority class by using the resampling method, as well as to improve the performance of traditional machine learning methods with the BRF approach. The findings obtained using the open source home credit dataset were evaluated using different performance criteria. Different applications using the resampling method for the unbalanced home credit data set show that BRF approach achieves superior classification accuracy compared to classical machine laarning methods.