An end-to-end AI-powered credit risk intelligence platform designed to simulate real-world banking and NBFC loan decision workflows using explainable machine learning, fairness monitoring, and interactive risk analytics.
This project predicts loan default probability, generates business-friendly approval decisions, explains model behavior using SHAP, and monitors fairness metrics for responsible AI governance.
Modern banking systems require more than just prediction accuracy.
Financial institutions must ensure:
- transparent decision-making
- explainable AI outputs
- fairness monitoring
- governance compliance
- operational risk control
- customer trust
This project demonstrates how AI can support:
- loan approval automation
- credit risk intelligence
- manual review prioritization
- explainable lending decisions
- fairness & bias monitoring
- responsible AI governance
- Predict probability of loan default
- Automate credit approval workflows
- Generate business-friendly decision explanations
- Monitor fairness across demographic groups
- Provide explainable AI reasoning using SHAP
- Simulate enterprise-grade credit risk systems
- Support governance and compliance visibility
Shows a low-risk applicant automatically approved by the AI system.
Displays a medium-risk customer routed for analyst review.
Shows a rejected applicant with transparent reason codes.
Approval-rate parity analysis for responsible AI governance.
Banks and fintech lenders constantly balance:
- growth
- customer acquisition
- portfolio quality
- credit risk
- regulatory compliance
Approving risky borrowers increases financial losses.
Rejecting too many safe borrowers reduces revenue growth.
This system helps answer:
- Should this loan be approved?
- Should it be sent for manual review?
- Why was this decision made?
- Is the model behaving fairly?
- Are governance thresholds being violated?
- Logistic Regression risk prediction
- probability of default estimation
- customer-level risk scoring
- threshold-based decisioning
- regulator-friendly interpretable modeling
Low-risk applicants are automatically approved.
Medium-risk applications are routed to analysts.
High-risk applicants are rejected with transparent reason codes.
- SHAP explainability integration
- feature contribution analysis
- adverse action reasoning
- interpretable model behavior
- transparent AI decision support
- approval-rate parity analysis
- gender fairness comparison
- governance threshold monitoring
- responsible AI auditing
- ethical AI system simulation
The dashboard includes:
- applicant simulation controls
- approval probability estimation
- manual review workflows
- rejection explanation engine
- fairness monitoring analytics
- real-time decision intelligence
Evaluation metrics used:
- Accuracy
- Precision
- Recall
- F1-Score
- ROC-AUC
The system prioritizes:
- interpretability
- governance readiness
- explainability
- operational usability
- responsible AI principles
This platform demonstrates how AI can help financial institutions:
- Reduce high-risk loan approvals
- Improve portfolio quality
- Accelerate lending operations
- Increase transparency in AI decisions
- Support compliance and governance teams
- Enhance customer trust through explainability
- Assist analysts with manual review prioritization
- Python
- Streamlit
- Scikit-learn
- Logistic Regression
- SHAP
- Fairlearn
- Pandas
- NumPy
- Matplotlib
The platform simulates a real-world banking AI risk pipeline:
- Applicant financial data is collected
- Features are engineered for credit risk modeling
- AI model predicts probability of default
- Decision engine classifies:
- Approved
- Manual Review
- Rejected
- SHAP explainability generates transparent reason codes
- Fairness monitoring evaluates governance metrics
- Results are visualized through the Streamlit dashboard
Explainable-Credit-Risk-Scoring/
│
├── data/
│ ├── raw/
│ │ └── default of credit card clients.xls
│ │
│ └── processed/
│ ├── cleaned_data.csv
│ └── features.csv
│
├── dashboard/
│ └── app.py
│
├── src/
│ ├── data_preprocessing.py
│ ├── feature_engineering.py
│ ├── train_model.py
│ ├── evaluate_model.py
│ ├── explain_model.py
│ └── fairness_analysis.py
│
├── models/
│ ├── credit_model.pkl
│ ├── scaler.pkl
│ └── feature_columns.pkl
│
├── outputs/
│ └── visuals/
│ ├── confusion_matrix.png
│ ├── roc_curve.png
│ ├── shap_summary.png
│ └── fairness_comparison.png
│
├── screenshots/
│ ├── loan_approval_dashboard.png
│ ├── manual_review_case.png
│ ├── loan_rejection_analysis.png
│ └── fairness_monitoring.png
│
├── reports/
│ ├── Model_Performance.md
│ ├── Explainability.md
│ └── Fairness_Bias.md
│
├── requirements.txt
├── .gitignore
└── README.mdThis project uses the well-known credit default dataset widely used in banking risk analytics and machine learning research.
- Credit card customer records
- repayment history
- bill statement information
- payment behavior analytics
- demographic variables
- default classification target
1→ Default Risk0→ Non-Default
git clone https://github.com/girishshenoy16/Explainable-Credit-Risk-Scoring.git
cd Explainable-Credit-Risk-Scoringpython -m venv venv
venv\Scripts\activatepython3 -m venv venv
source venv/bin/activatepip install --upgrade pip
pip install -r requirements.txtpython src/data_preprocessing.py
python src/feature_engineering.py
python src/train_model.py
python src/evaluate_model.py
python src/explain_model.py
python src/fairness_analysis.pystreamlit run dashboard/app.pyThe platform generates:
- approval probability analysis
- loan decision outputs
- fairness comparison reports
- SHAP explainability visuals
- confusion matrices
- ROC curve analytics
- governance monitoring outputs
This project demonstrates responsible AI concepts commonly used in regulated financial systems:
- Explainable AI for transparent lending decisions
- Fairness monitoring across demographic groups
- Human-in-the-loop manual review workflows
- Governance-oriented risk thresholds
- Interpretable machine learning models
- Bias awareness and monitoring simulation
Note: This project is an educational simulation and does not represent a production banking system.
Modern AI systems in banking cannot function as “black boxes.”
Financial institutions increasingly require:
- explainable AI
- responsible lending systems
- fairness auditing
- governance monitoring
- transparent decision intelligence
This project demonstrates how AI can move beyond prediction into:
- operational decision systems
- explainable banking intelligence
- ethical AI governance
- enterprise-grade financial analytics
Girish Shenoy
Aspiring AI & Data Analytics Professional focused on:
- Explainable AI
- Financial Analytics
- Risk Intelligence
- AI Governance
- Machine Learning Systems
- Decision Intelligence Platforms
Guided by Umesh Yadav Sir under EDC, IIT Delhi in association with the Indian Institute of Placement.