Predict default of credit card clients

This dataset from the UCI Machine learning repository [1] represent the default payment from Taiwan custumers. We are aimed to predict the payment based on independet variables from the custumers.

Exploratory Data Analisys

We can see that the No Payment group has a smooth histogram that does lead to a visible pattern. On the other hand, the Payment group shows a fat tail, indicating that an older popullation is more representative in this group.

As we can expected, lower credit amounts are more likely to be paid, while higher credit amounts are less likely to be paid.

Just by looking at the proportion, there is a slightly higher probability that a male customer will pay the credit, but it is not a statistically significant difference.

Unexpectedly, the No Payment group has a higher population of individuals from Graduate School.

The Payment group has a slightly higher population of married customers, but again, it is not statistically significant

The negative status represents timely payment, while the positive status indicates a delay in payment, each number representing the amount of months of delay. Notably, the No Payment group tends to delay their payments.

As expected, the higher the bill amount, the less likely it is to be paid.

No surprise from the previous distribution, payments tend to be shorter in the Payment group

Model

Accuracy: 0.72

Accuracy: 0.82

In the context of this particular problem, a false positive incurs significant costs for the Credit Card Company. Our baseline model achieves an accuracy of 0.73 but results in many false positives. The Random Forest model shows a significant improvement, achieving an accuracy of 0.81 and producing far fewer false positives. Lastly, the Neural Network approach does not outperform as expected, but it provides a solid accuracy of 0.81, similar to the Random Forest approach.

We have successfully trained a useful and robust model to predict customer default payments

Thanks for reading! 🧑‍💻💕