A Machine Learning Framework for Improving Classification Performance on Credit Approval

Credit scoring is a model commonly used in the decision-making process to refuse or accept loan requests. The credit score model depends on the type of loan or credit and is complemented by various credit factors. At present, there is no accurate model for determining which creditors are eligible for loans. Therefore, an accurate and automatic model is needed to make it easier for banks to determine appropriate creditors. To address the problem, we propose a new approach using the combination of a machine learning algorithm (Naïve Bayes), Information Gain (IG), and discretization in classifying creditors. This research work employed an experimental method using the Weka application. Australian Credit Approval data was used as a dataset, which contains 690 instances of data. In this study, Information Gain is employed as a feature selection to select relevant features so that the Naïve Bayes algorithm can work optimally. The confusion matrix is used as an evaluator and 10-fold cross-validation as a validator. Based on experimental results, our proposed method could improve the classification performance, which reached the highest performance in average accuracy, precision, recall, and f-measure with the value of 86.29%, 86.33%, 86.29%, 86.30%, and 91.52%, respectively. Besides, the proposed method also obtains 91.52% of the ROC area. It indicates that our proposed method can be classified as an excellent classification.