Credit Card Approval Prediction Machine Learning
Credit cards have become one of the most popular modes of payment in the world today. It has therefore become pertinent for issuers of credit cards to identify the risks involved or the probability of the consumers failing to meet their obligations thereby incurring costs. To mitigate this risk, financial institutions utilise efficient credit risk evaluation tools such as credit scores which are based on data on the potential customers’ sociodemographic status, credit reports and other criteria, before issuing credit. A credit scoring system reduces human errors and increases the accuracy and speed of credit risk determination which in turn reduces the time and process of granting loans, the costs and risks of granting loans, and increases the efficiency and transparency of the bank. The purpose of the credit scoring system, therefore, is to announce the quality of the credit customer requesting the loan and to predict their repayment behaviour. The objective of this report is to develop and critically evaluate the machine learning models to predict the creditworthiness of credit card applicants, compare models based on certain performance metrics such as their accuracy, sensitivity and specificity and to devise a business strategy for profitability. In the case where the target feature to be predicted is whether an applicant will default on a loan, the supervised learning algorithms are mostly used, and fall under the classification models. A wide range of classification methods such as Bayesian Network, Naïve Bayes, Support Vector Machine (SVM), Logistic Regression, K-Nearest Neighbour, Decision Tree, Random Forest, Neural Network etc. are frequently used to detect financial risks such as the credit worthiness of loan applicants (Peng et al., 2011). The performance of the classifier may vary based on the data, performance measure and other circumstances therefore the selection of a suitable classifier remains an important task in the prediction of financial risk.