As a machine learning enthusiast, I was thrilled to see a project that combined the power of XGBoost, SHAP, and Streamlit to build an end-to-end credit risk model. The project’s goal was to predict loan defaults with high accuracy and provide explainable results. Let’s dive into the details.
The model achieved impressive results, with a ROC AUC of 0.73 and a recall of 76% for catching defaults. The business-optimized threshold resulted in a 50% approval rate and a 9.7% bad rate. But what really stood out was the use of SHAP explanations for every loan decision, providing transparency and insights into the model’s predictions.
The project’s production-readiness was also noteworthy, with modular .py scripts and an interactive Streamlit dashboard. This makes it easy to deploy and maintain the model in a real-world setting.
If you’re interested in learning more, the project is open-sourced on GitHub, and I encourage you to check it out.
Overall, this project demonstrates the potential of machine learning in credit risk assessment and sets a high standard for future projects in this domain.