Problem Statement

In this case study, we worked with a fintech company to help them predict the Potential Defaulters and Non-Defaulters among the customers who are applying vehicle loan to help Employers to decide whether to provide customers loan or not.

Approach

Basic univariate analysis is done to understand the data and the the distribution of the data. Data set consists of the vehicle cost, dealer details and the customer details and the outcome of the loan defaulted.

The Outcome of the defaulter is imbalanced where the 78.12 21.7 split. SMOTE analysis is done to deal with the class imbalance.

Feature engineering is done to get derived calculation such as CIBIL score of the customer, the frequency of payments and the proof details.

All base classification models are used to check the F1 score of the model performance. Ensemble model XGboost is selected as the final model to find the people who are defaulting their payments.

Conclusion

The model gives great validation scores and is significantly better than another model done on this subject. Due to this, we projected about 16.7% growth in profit by accurately filtering out rightful defaulters.