Performance Evaluation of Machine Learning Approaches for Credit Scoring

Abstract

In today’s world, assessing financial credit risk is of immense importance in both accounting and finance areas. Financial institutions need to keep the credit default risk to an acceptable level so that higher profitability can be achieved. Recently, with the fast development of modern data science, many machine learning methods have been applied to make accurate predictions based on the information extracted from diverse data sources. The present study aims to apply data mining techniques in acquiring evidence used to judge which classifier performs better in assessing credit scoring for a proposed model. The two datasets employed in the analysis of this paper are the “Give Me Some Credit” dataset and the “PPDai” dataset. Eight classification methods are adopted in the paper including Linear Discriminant Analysis (LDA), Logistic Regression (LR), Decision Tree (DT), Support Vector Machine (SVM), Random Forest (RF), Gradient Boosting Decision Tree (GBDT), eXtreme Gradient Boosting (XGboost) and Multi-Layer Perceptron (MLP). Three indicators (Accuracy, AUC and Logistic loss) are used to analyze the performance of each classifier. The final experiment results indicate that the XGBoost classifier has a better performance in predictive analytics compared with the other seven models. The study results will also provide practical values for financial institutions in choosing the appropriate classifier so as to make correct judgements when they are faced with credit problems in real situations.