Logo Lanfrica

highfrezh/fairlend-africa

Domain:

socioeconomic

Record type:

software
Creator:
hig
Host:
This is a Research Demonstration Project developed for academic transparency and to showcase high-stakes AI auditing. It is not intended for commercial use. # 🌍 FairLend-Africa **Explainable AI for Alternative Credit Scoring in Financially Excluded Communities** --- ## 📌 Overview Access to formal credit is a fundamental barrier to financial inclusion in Sub-Saharan Africa. **FairLend-Africa** is a research-grade framework that proves how behavioral data—mobile money transactions, savings consistency, and airtime habits—can serve as powerful, fair, and explainable proxies for creditworthiness. > [!NOTE] > This is a **Research Demonstration Project** developed for academic transparency and to showcase high-stakes AI auditing. It is not intended for commercial use. ### 🧪 Research at a Glance | Metric | Performance | | :--- | :--- | | **Model Ranking (ROC-AUC)** | **0.7137** (Peer-benchmarked) | | **Decision Precision** | **86.0%** | | **Fairness Compliance** | **100%** (Zero violations of the 80% rule) | | **Primary Signal** | `wallet_balance_trend` (SHAP: 0.377) | --- ## 🖥️ Dashboard Preview *The FairLend Dashboard provides loan officers with real-time SHAP-based explanations for every credit decision, bridging the gap between "Black Box" AI and human understanding.* --- ## 🏗️ Project Architecture ```text fairlend-africa/ ├── notebooks/ # End-to-end research pipeline (01-06) ├── api/ # High-performance FastAPI backend ├── frontend/ # Interactive React dashboard ├── src/ # Core logic for data & ML auditing ├── artifacts/ # Generated plots, metrics, and serialized models └── paper/ # 📄 fairlend_africa.pdf (Manuscript) ``` --- ## 🚀 Quick Start (Reproduction Guide) ### 1. Environment Setup ```bash git clone github.com cd fairlend-africa python -m venv .venv source .venv/bin/activate # Windows: .venv\Scripts\activate pip install -r requirements.txt ``` ### 2. Generate Data & Train Model ```bash # Generate 10,000 synthetic behavioral records python src/data/generate_dataset.py --n 10000 # Execute the pipeline (o …

Licenses