Credit Card Approval Predictions Using Logistic Regression, Linear SVM and Naïve Bayes Classifier

With the huge growth of financial institution databases, the evaluation of credit issuance decisions begins to improve the decision-making methods of manual judgment and statistical analysis, which greatly improves the reliability and efficiency of credit issuance decisions. Machine learning algorithms, as one of the most important statistical tools, have witnessed increasing importance in supporting credit approval decisions. However, based on the different algorithms of each model and the selection of corresponding parameters in a given model, the prediction performances have varied among prediction models. To better evaluate the prediction performance of prevalent models and improve the model construction in the credit scoring process, this paper analyzes the prediction accuracy of multiple regression models and classifiers based on a predetermined performance criterion and proposes an optimal model with the highest prediction accuracy. The experimental models involved in the analysis include Logistic Regression, Linear Support Vector Classification (Linear SVC) and Naïve Bayes Classifier. According to the reported results, Linear SVC performed well in the analyzed model, reflecting the highest performance score of Balanced Accuracy (89.09%) and Accuracy Rate (88.48%).