Credit Risk Analysis & Prediction Model

Minimizing Financial Default using Machine Learning

Project Focus: Banking Strategy | Risk Management | Predictive Analytics

---

Executive Summary

In the banking sector, the cost of a "False Negative" (approving a bad loan) is significantly higher than a "False Positive" (rejecting a good loan).

This project analyzes German credit data to build a scoring model that prioritizes Capital Preservation. By adjusting the algorithm's sensitivity, I successfully increased the detection of potential defaulters (Recall) from 27% to 71%, significantly reducing the bank's exposure to toxic debt.

---

Key Business Insights

1. The "Danger Zone": Analysis revealed a high concentration of default risk in applicants under 30 years old requesting credits above 5,000 DM.
2. Risk Calibration: The initial model had high accuracy (72%) but failed to detect risk. The optimized model sacrifices marginal accuracy (68%) to achieve a 2.5x improvement in risk detection.

Technical Implementation

• Data Cleaning: Handled missing financial records and mapped German legacy terms to Business English standard variables.
• Visualization: Created scatter plots (Age vs. Amount) to identify risk clusters visually.
• Machine Learning:
  • Algorithm: Logistic Regression (Chosen for interpretability in a regulated sector).
  • Optimization: Applied StandardScaler and Class Weight Balancing to correct bias towards the majority class.

Model Performance (Confusion Matrix)

Metric	Value	Business Impact
Accuracy	68%	Overall correctness of the system.
Recall (Bad Credit)	71%	Critical Metric: Percentage of actual defaulters correctly identified and rejected.
Precision	49%	Accepted trade-off to ensure maximum safety.

---

How to Run This Project

1. Clone the repository.
2. Install dependencies: pandas, seaborn, scikit-learn.
3. Run Credit_Risk_Scoring_Model.ipynb to see the training process and visualizations.

---

"Banking is not about avoiding risks, but about managing them intelligently."