Credit Card Default Prediction and Analysis
The purpose of this project is to present the analysis and study of default on credit cards based on the demographics, credit data, history of payment and periodic bill statements of the credit card clients in Taiwan. This study will help in identifying the potential defaulters for a credit card company so that the company can handle the situation in advance to deal with the clients showing higher risks. It is helpful for creditors to assess the future standing of a client to avoid serious credit card statuses like ‘severely delinquent’ on the credit card payment and any new credit obligations. Simultaneously, companies can leverage this analysis in identifying their existing ideal clients or potential future customers.
When it comes to identifying the default on credit cards, the history of the past few months play a key role. We have done statistical analysis on our dataset which consists of bill statement, amount of previous payment, repayment status and balance limit to name a few. Default on credit cards is affected by education level, marriage status, age and sex of the client along with the payment history, considering this, we have used supervised machine learning techniques to identify the potential defaulters. We have constructed our models using classification techniques such as Logistic Regression, Kernel SVM, Random Forest, XGBoosting and Gradient-Boosting to classify the credit card clients into two groups, defaulters and non-defaulters. For evaluating the robustness of our model, we have used k-fold cross-validation and since our dataset exhibits class imbalance, we have used F-measure as the metric to identify a trusting and well-performing model.