A simple classification problem based on credit score data which allows us to identify whether a particular loan applicant may be given or denied credit (loan). Using Rattle and R (for some box-plot snippets), we've tried to bring out some interesting insights.
Apart from studying the descriptive statistics of a few of the explanatory variables, we've also tried to fit in a few probabilistic models like the Logistic Regression model, Adaptive Boosting model, Support Vector model and Decision Trees.
Our understanding of the project, given the classification point of view, is that Decision Trees never should be used to depict a credit scoring classification problem. It just does not make any sense! Citing an example, having a checking account (equivalent as having a current account in India) with more than substantial balance and not having such an account at all, are both likely conditions for a borrower to be classified as a good credit. As a result of which we have decided not to include the decision tree into our presentation apart from its score which reflects a much higher error in terms of the other models used.
Most of the business implications have been included in the presentation itself.
Our conclusion is that we may try to project a model, based on the average scores of the models already cited in the presentation, instead of recording each representative model score. This approach would further reduce the error rate.
Please click here to view the presentation.
Cheers!
