This project describes the experimentation conducted with data collected over years for deciding whether a potential customer would default from paying back the load. The work describes appraoches based on machine learning for predicting the status of a loan. Different machine learning approaches are used in this study and AUC-ROC is considered as the metric of evaluation.
The approaches used for experimentation and metrics:
| Algorithm | AUC-ROC |
|---|---|
| Logistic Regression(LR) | 50.56 |
| Boosted Logistic Regression | 59.61 |
| Decision Tree | 73.45 |
| Random Forest(RF) | 73.45 |
| K-Nearest Neighbors(KNN) | 61.85 |
| Ensemble (LR,RF,KNN) | 62.7 |
To run the code clone the project
- train.py is used for training - path to training data is required
- test.py is used for inference - path to test data and trained model file required
Trained models for the above experiments are available at: https://drive.google.com/file/d/1HLquUEZ0iQDWMneXlTP5gZ54MSmFhxX9/view?usp=sharing