AlmaBetter Verfied Project - AlmaBetter School
This project is aimed at predicting the case of customers default payments in Taiwan.
From the perspective of risk management, the result of predictive accuracy of the estimated probability of default will be more valuable than the binary result of classification; 'credible' or 'not credible' clients.
The project has been conducted in 5 steps:
- Data Cleaning
- Exploratory Data Analysis (EDA)
- Data Transformation
- Model Building and Evaluation
- Hyperparameter Tuning
Based on following ML-classification models:-
- After performing the various model we get the best accuracy form the all the models except random forest.
- Random Forest is the least accurate as compared to other models as it shows overfitting of the model.
- XG boost and Logistic regression has the best precision and recall Balance.
- Higher recall can be achieved if low precision is acceptable.
- We can deploy the model and can be served as an aid to human decision.
- Model can be improved with more data and computational resources.
- It is important to have models deployed on the cloud with live customer interaction through app or mobile phone so that suspicious activities/a missed payment can be verified real-time with a text message or a push notification.