Comparative Studies on Methods of Selecting Features in Credit Scoring

Feature selection (FS) is a critical process in credit scoring, which can improve models, address overfitting, decrease computational cost, and increase model interpretability. This paper presents a classification of research problems and methods related to the above issues. The research problems can be divided into two types based on focus: model development or evaluation, and two types based on methodology: machine learning or operations research. The machine learning research objects can be divided into four types based on the feature of training and type of optimization. The paper also provides system factors and evaluation metrics of the experiments for FS in credit scoring. The system factors include personal information, credit history, type of loan, loan amount, and loan purpose. The evaluation metrics include accuracy, precision, recall, F1 score, and AUC.