Comparative study of classical machine learning models in credit scoring

Credit scoring is a crucial aspect of financial risk management, enabling lenders to evaluate the creditworthiness of individ uals and entities. This paper presents a comprehensive comparison of various classical machine learning models to determine the most effective one for predicting credit scores. The evaluated models include logistic regression, decision trees, random forests, support vector machines, gradient boosting, adaptive boosting (AdaBoost), k-nearest neighbors (KNN), Naive Bayes, eXtreme Gradient Boosting (XGBoost), Light Gradient-Boosting Machine (LightGBM), multi-layer perceptron (MLP) classifiers, deep neural networks (DNN), and convolutional neural networks (CNN). Each model underwent 10-fold cross-validation and was assessed based on metrics such as accuracy, precision, recall, F1 score, and receiver operating characteristic curve (ROC curve). Results indicate that XGBoost and LightGBM outperform other models across multiple metrics, demonstrating their suitability for credit score prediction. The paper concludes with recommendations for selecting models tailored to specific business requirements and data characteristics, highlighting the significant potential of advanced machine learning techniques in enhancing credit scoring accuracy and reliability.