PREDICTING BANK LOAN APPROVAL USING LOGISTIC REGRESSION AND FEATURE SELECTION METHOD

Abstract

Examining bank loan application is a long process that requires a detail check in every stages. This process is important in the banking industry, as it directly impacts the bank's risk management and profitability. However, due to a long process in making decision the customer wait a long time to get the decision which results to the customers dissatisfaction. Therefore, to improve the examination process and provide a quick decision result, the more effective tools is required. Logistic regression is a machine learning method that able to predict the binary output based on the probability value. This method takes the value from the multiple regression method and convert it into probability value using the activation function called sigmoid function. This paper applies the logistic regression method to predict bank loan approvals based on several features considered as independent variables. This research uses the secondary data taken from www.kaggle.com. The model performance is measured using the confusion matrix that consist of accuracy, precision, recall, and F1 score. This research construct three models based on data type. The first model is constructed using numerical data only, the second model is constructed using categorical data only, and the third model is constructed by combining numerical data type and categorical data type. It is determined that the first model return 87.8% accuracy, 95.94% precision, 87.57% recall, and 91.56% F1 score. The second model return accuracy 69%, precision 96.81%, recall 69.87%, and F1 score 81.17%. Moreover, the return 86.6% accuracy, 96.81% precision, 85.64% recall, and 90.88% F1 score. Based on these results, it is concluded that the best method in processing loan bank application data is to use the third model that is the model that includes both categorical and numerical data type.