Built for the AI Tool Development Challenge 2026 — One Planet. One Purpose. Powered by AI.
Across South Asia, an estimated 300–400 million workers operate in the informal economy — street vendors, daily wage labourers, gig delivery workers, auto-rickshaw drivers, and small shopkeepers. Despite contributing significantly to national GDP, they are systematically excluded from formal financial systems because they have no traditional financial records — no payslips, no bank statements, no credit history.
Without these, they cannot access:
- Formal bank loans or micro-credit
- Government welfare and subsidy programs
- Insurance or emergency financial safety nets
FinInclusion AI solves this by using alternative economic indicators — income patterns, transaction frequency, digital payment behaviour — to assess financial stability and connect workers to the right support.
| Priority | SDG | Relevance |
|---|---|---|
| 🥇 Primary | SDG 16 — Governance, Transparency & Justice | Explainable AI decisions for fair, accountable credit assessment |
| 🥈 Core | SDG 1 — No Poverty | Connecting excluded workers to credit and welfare |
| 🥈 Core | SDG 10 — Reduced Inequalities | Alternative scoring levels the playing field |
| 🥉 Supporting | SDG 8 — Decent Work & Economic Growth | Enabling micro-credit for entrepreneurship |
| 🥉 Supporting | SDG 11 — Sustainable Cities & Communities | Targeting urban informal workers |
| 🥉 Supporting | SDG 9 — Smart Industry & Infrastructure | Scalable AI financial infrastructure |
| ➕ Additional | SDG 17 — Partnerships for the Goals | Designed for MFI, NGO, and government partnerships |
Instead of asking "What is your bank balance?", the system asks:
"How consistently do you earn? How often do you transact? How digitally active are you?"
It builds a financial profile from behaviour rather than formal records, and outputs:
| Output | Description |
|---|---|
| 🏷️ Stability Label | Stable / Moderate / Unstable |
| 📊 Credit Score | Alternative score from 300 to 850 |
| Low / Medium / High | |
| 🏛️ Support Program | Micro-Investment / Micro-Credit / Government Welfare |
| 🔍 Factor Analysis | Which inputs helped or hurt the score — and by how much |
finclusion/
│
├── data/
│ ├── generate_dataset.py ← Generates 5,000 synthetic worker profiles
│ ├── raw/
│ │ └── informal_workers.csv ← Raw synthetic dataset (auto-created)
│ └── processed/
│ └── processed_workers.csv ← Cleaned, encoded, normalised dataset
│
├── models/
│ └── financial_model.pkl ← Trained GradientBoosting model (auto-saved)
│
├── utils/
│ ├── preprocess.py ← Full preprocessing pipeline
│ └── model.py ← ML training, evaluation & prediction
│
├── app/
│ └── main.py ← Streamlit web app (entry point)
│
|
│-----Project-Overview-document-Google-docs <------------------(Simplified Document)
│----Project-Overview-updated <-----------(Iilustrated Document)
├── requirements.txt ← All Python dependencies
└── README.md
- Python 3.9 or higher
- pip
cd finclusion# Create
python -m venv venv
# Activate on Windows
venv\Scripts\activate
# Activate on macOS / Linux
source venv/bin/activatepip install -r requirements.txtpython -c "import pandas, numpy, sklearn, joblib, streamlit, matplotlib; print('All libraries OK ✅')"Run these commands in order from inside the finclusion/ folder:
python data/generate_dataset.pyCreates data/raw/informal_workers.csv with 5,000 worker profiles.
python utils/preprocess.pyCleans, engineers features, encodes, and normalises. Saves to data/processed/processed_workers.csv.
python utils/model.pyTrains Random Forest and Gradient Boosting, auto-selects the best, saves to models/financial_model.pkl.
# Recommended on Windows
python -m streamlit run app/main.py
# If port 8501 is blocked, use a different port
python -m streamlit run app/main.py --server.port 8502Opens in your browser at http://localhost:8501
| Field | Type | Range |
|---|---|---|
| Occupation | Dropdown | Street Vendor, Driver, Laborer, Delivery Worker, Shopkeeper |
| Daily Income (₹) | Slider | ₹200 — ₹2,000 |
| Work Days / Month | Slider | 1 — 30 |
| Monthly Transactions | Slider | 1 — 120 |
| Savings Ratio | Slider | 0.00 — 0.40 |
| Digital Payment Ratio | Slider | 0.00 — 1.00 |
| Loan History | Radio | Yes / No |
- 4 Metric Cards — Stability Label, Credit Score, Risk Level, Model Confidence
- Credit Score Gauge — Semicircle needle chart (red → orange → green)
- Global Feature Importance — Which features matter most across all workers
- Factor Contribution Analysis — Waterfall chart showing exactly why this specific worker got their prediction (🟢 green = helps score, 🔴 red = hurts score)
- Plain-English Summary — One sentence explaining the prediction in simple language
- Probability Breakdown — Unstable / Moderate / Stable class probabilities
- Support Program Recommendation — Tailored welfare or credit program
| Model | Test Accuracy | Selected |
|---|---|---|
| RandomForestClassifier | 89.9% | |
| GradientBoostingClassifier | 90.5% | ✅ Auto-selected |
| Feature | What It Measures |
|---|---|
daily_income |
Average earnings per working day |
work_days_per_month |
Employment consistency |
monthly_transactions |
Economic activity level |
savings_ratio |
Financial discipline |
income_min / income_max |
Income range |
digital_payment_ratio |
Financial inclusion level |
loan_history |
Existing credit exposure (binary) |
income_consistency |
Engineered — min/max income ratio |
monthly_income |
Engineered — daily income × work days |
| Label | Meaning | Credit Score Range |
|---|---|---|
| Stable | Financially resilient | 700 — 850 |
| Moderate | Some vulnerability | 500 — 699 |
| Unstable | High financial risk | 300 — 499 |
stability_score = P(Stable) × 1.0 + P(Moderate) × 0.5 + P(Unstable) × 0.0
credit_score = 300 + (stability_score × 550)
| Class | Precision | Recall | F1-Score |
|---|---|---|---|
| Unstable | 0.93 | 0.92 | 0.92 |
| Moderate | 0.90 | 0.91 | 0.90 |
| Stable | 0.89 | 0.87 | 0.88 |
| Overall | 0.91 | 0.91 | 0.90 |
The synthetic dataset contains 5,000 rows generated with realistic occupation-based profiles:
| Occupation | Income Range | Work Days | Digital Ratio |
|---|---|---|---|
| Street Vendor | ₹200 – ₹900 | ~24/month | Low (5–35%) |
| Driver | ₹500 – ₹1,600 | ~22/month | Medium (30–80%) |
| Laborer | ₹200 – ₹700 | ~20/month | Very Low (2–20%) |
| Delivery Worker | ₹600 – ₹1,800 | ~25/month | High (50–95%) |
| Shopkeeper | ₹700 – ₹2,000 | ~26/month | Medium (25–75%) |
Label distribution: Moderate (49%) · Unstable (31%) · Stable (20%)
| Library | Version | Purpose |
|---|---|---|
pandas |
2.2.2 | Data loading, cleaning, manipulation |
numpy |
1.26.4 | Numerical operations |
scikit-learn |
1.5.0 | ML models, preprocessing, evaluation |
joblib |
1.4.2 | Save and load trained models |
streamlit |
1.35.0 | Interactive web prototype |
matplotlib |
3.9.0 | Charts and visualisations |
seaborn |
0.13.2 | Statistical plots |
| Phase | Status | Description |
|---|---|---|
| POC | ✅ Complete | Synthetic data, trained model, Streamlit prototype, explainability |
| Pilot | 🔜 Next | Partner with MFI/NGO, collect real worker data |
| Regional Scale | 📋 Planned | Multilingual UI (Hindi, Urdu, Bengali), REST API, mobile PWA |
| National Scale | 📋 Planned | Federated learning, Aadhaar integration, cloud deployment |
- Transparency — Every prediction includes a Factor Contribution Analysis explaining the decision
- Fairness — Model audited for bias across occupation groups and income levels
- Privacy — No raw personal data stored after prediction
- Accountability — Workers can request re-assessment after improving financial behaviour
- Consent-first — Workers explicitly agree to data use before assessment
| KPI | Current (POC) | Target (Production) |
|---|---|---|
| Model Accuracy | 90.5% | ≥ 88% on real data |
| Recall — Unstable class | 92% | ≥ 90% |
| F1 Score (Macro Average) | 0.90 | ≥ 0.87 |
| Workers Assessed / year | — | 10,000 |
| Micro-credit Approvals Enabled | — | 2,000 |
| Welfare Enrollments | — | 1,500 |
| False Denial Rate | — | < 8% |
| Prediction | Recommended Program | Max Support |
|---|---|---|
| 🟢 Stable | Micro-Investment Scheme | ₹50,000 loan at subsidised rate |
| 🟡 Moderate | Micro-Credit + Skills Training | ₹20,000 + free vocational training |
| 🔴 Unstable | Government Welfare Support | PM Jan Dhan Yojana + emergency relief |
- Informal Workers — Get a credit score and support recommendation without formal records
- Microfinance Institutions — Automate informal applicant screening
- Government Welfare Agencies — AI-assisted welfare eligibility identification
- NGOs & Field Workers — Rapid on-the-spot assessment tool
- Policymakers & Researchers — Aggregate insights on financial vulnerability patterns
Full submission documentation is in FinInclusion_Project_Overview.docx covering:
- Problem statement and SDG alignment in depth
- System architecture and end-to-end workflow
- Expected impact and real-world use cases
- Scalability and deployment roadmap
- Full KPI framework
This project is licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0).
You are free to:
- Share — copy and redistribute the material in any medium or format
- Adapt — remix, transform, and build upon the material for any purpose, even commercially
Under the following terms:
- Attribution — You must give appropriate credit, provide a link to the license, and indicate if changes were made.
🔗 Full license text: https://creativecommons.org/licenses/by/4.0/
Built with ❤️ for financial inclusion AI Tool Development Challenge 2026 — One Planet. One Purpose. Powered by AI. UN SDG Alignment: 16 · 1 · 10 · 8 · 11 · 9 · 17