🏦 Bank Loan Repayment Analysis

Course: DAMO-611-2 Data Analytics Case Study 3 (Spring 2025)
Instructor: Omid Isfahanialamdari
Submitted by:

• Heta Chavda (NF1014555)
• Yash Patel (NF1009944)
• Devarsh Oza (NF1003776)
• Joy Ajayi (NF1002698)

---

📌 Project Overview

Banks face increasing challenges with loan defaults (non-performing loans), which affect profitability, compliance, and investor confidence.
This project leverages Logistic Regression to predict loan repayment outcomes (“Fully Paid” vs. “Charged Off”) using borrower and loan-specific features.

Key Goals:

• Reduce default risk via predictive analytics.
• Provide actionable recommendations for financial institutions.
• Build a transparent, interpretable model suitable for deployment.

---

🎯 Objectives

• Classify loans as “Fully Paid” or “Charged Off” using borrower attributes.
• Quantify the impact of key variables:
  • Interest Rate
  • Annual Income
  • Debt-to-Income (DTI) ratio
  • Employment Length
• Support business decision-making with risk segmentation.
• Provide a framework for deployment & monitoring:contentReference[oaicite:0]{index=0}:contentReference[oaicite:1]{index=1}.

---

📂 Dataset & Preprocessing

• Source: Internal lending dataset (financial_loan.csv)
• Rows: ~37,138 loan applications
• Target Variable: loan_status → Binarized into Paid (0) and Charged Off (1)

Preprocessing Steps

• Removed missing records in critical fields.
• Encoded emp_length → Numeric scale (0–10).
• Scaled continuous variables (int_rate, dti, income).
• Created binary classification target.

---

🔎 Exploratory Data Analysis

Loan Status Distribution

• 84% fully repaid
• 14% charged off
• 3% current loans:contentReference[oaicite:2]{index=2}

---

Loan Status by Employment Length

• Longer tenure (10+ years) = highest repayment likelihood.
• Defaults remain consistent (~12–15%) across groups:contentReference[oaicite:3]{index=3}.

---

DTI vs Loan Status

• Median DTI:
  • Charged Off: ~14.5%
  • Fully Paid: ~13%
• Higher DTI correlates with repayment challenges:contentReference[oaicite:4]{index=4}.

---

Interest Rate vs Loan Status

• Higher interest rates → higher default probability.
• Fully repaid loans cluster at lower interest brackets:contentReference[oaicite:5]{index=5}.

---

🤖 Modeling Approach

Algorithm

• Logistic Regression (chosen for interpretability & statistical rigor).

Workflow

1. Train/Test Split: 80/20
2. Class Imbalance: class_weight='balanced'
3. Metrics: ROC-AUC, Precision, Recall, F1, Confusion Matrix:contentReference[oaicite:6]{index=6}

---

📊 Model Performance

• ROC-AUC Score: 0.9154 → Strong predictive performance:contentReference[oaicite:7]{index=7}
• Confusion Matrix Results:
  • TP = 3922
  • TN = 768
  • FP = 477
  • FN = 2261

Interpretation

• Interest Rate (–16.45): Strong negative driver (higher rates → higher defaults).
• DTI (–1.06): Higher ratios → more defaults.
• Employment Length (–0.03): Longer tenure slightly reduces default risk.
• Annual Income (+0.000003): Minimal positive effect:contentReference[oaicite:8]{index=8}.

---

✅ Business Impact

• Risk Mitigation: Early identification of high-risk borrowers.
• Portfolio Optimization: Shift focus toward safer borrowers.
• Automation: Integration with digital workflows enables real-time credit scoring.
• Customer Targeting: Personalized lending terms by borrower risk segment:contentReference[oaicite:9]{index=9}.

---

📌 Recommendations

1. Deploy Model in Pre-Loan Screening

   • Integrate logistic regression scoring into loan origination platforms.
   • Flag risky applicants with >70% default probability.

2. Retrain Quarterly

   • Capture market shifts and borrower behavior changes.

3. Expand Feature Set

   • Add credit bureau data, loan purpose, payment history.

4. Test Ensemble Models

   • Random Forest, Gradient Boosting (XGBoost) for higher accuracy.

5. Governance & Compliance

   • Ensure transparency, fairness checks, explainability (e.g., SHAP values).
   • Align with Basel III/IV & Fair Lending regulations:contentReference[oaicite:10]{index=10}:contentReference[oaicite:11]{index=11}.

📌 Conclusion

The Logistic Regression model demonstrated strong predictive power with ROC-AUC of 0.91, making it a reliable solution for loan risk prediction.
With ongoing retraining, feature expansion, and compliance oversight, the model can:

• Reduce defaults
• Improve profitability
• Strengthen trust with regulators and investors

---

📚 References

1. Chris Ekai (2023). Credit Risk Assessment: A Comprehensive Guide for Lenders and Financial Institutions. Risk Publishing.
2. Atul K. Gupta (2024). Credit Risk Predictive Modeling. LinkedIn Pulse.

---