Credit risk-type classification using statistical learning methods Available to Purchase

Classifying applicants with good credit risk is essential for developing the banking industry in the future. On the other hand, applicants with bad risk can certainly make banks, as financial institutions provide credit, experience difficulties. The opportunity for bad credit risk can be great if the credit application acceptance system misclassifies applicants with a bad risk. This paper aims to investigate two statistical learning models to classify applicants who have good credit risk reliably; multinomial naïve Bayes and decision tree ID3. These nonparametric models require that all predictor variables are of categorical type. The proposed strategy discretizes the numeric type variables using the Sturges frequency distribution table. The results, validated using k-fold cross-validation resampling methods, showed that the strategy is very appropriate, especially in the ID3 decision tree model, because it produces a classification with performance measures greater than 90% for accuracy and sensitivity, precision, and F1-score. Only the specificity is 83.23%. However, sensitivity as a performance measure that prioritizes credit risk has a value of almost 100%, namely 99.87%. The multinomial naive Bayes method also has good performance. All performance measures have a greater than 85% value, and only specificity is below 85%. These performance measures indicate that the ID3 decision tree method is better than the multinomial naive Bayes method for classifying credit risk.