Lateephah/Loan-Default-Risk-Predictor-Micro-Lenders-POS-Agents-Nigeria-

A tabular machine learning model that predicts the likelihood of loan default, built for Nigerian micro-lenders and POS (Point-of-Sale) agents who need a fast, explainable repayment-risk check before disbursing small, short-tenor loans. # Loan Default Risk Predictor for Micro-Lenders & POS Agents (Nigeria) A tabular machine learning model that predicts the likelihood of loan default, built for Nigerian micro-lenders and POS (Point-of-Sale) agents who need a fast, explainable repayment-risk check before disbursing small, short-tenor loans. **▶ Watch the 3-minute demo video** --- ## Table of Contents - Problem Statement - Solution Overview - Data - Methodology - Results - How to Run - Limitations & Future Work - Acknowledgments --- ## Problem Statement Across Nigeria, POS agents and micro-lenders extend small, short-term loans to informal traders, artisans, and small business owners often with little to no formal credit history to underwrite against. A wrong call is costly in either direction: lending to a high-risk borrower risks non-repayment, while rejecting a genuinely low-risk borrower loses income and excludes someone who may be perfectly creditworthy. This project builds a lightweight, explainable risk-scoring model that a micro-lender or POS agent could realistically use at the point of a lending decision, fed either a single applicant's details (manual input) or a batch of applicants (CSV upload). ## Solution Overview - **Input:** borrower demographic, business, POS-transaction, and loan-request data via manual entry or CSV upload - **Output:** a default-risk probability (0–100%), a risk band (Low / Medium / High), and the key factors behind the score - **Models compared:** Logistic Regression, Random Forest, and XGBoost, each evaluated under three class-imbalance strategies (baseline, class weighting, SMOTE), nine variants total - **Deployed model:** XGBoost with class weighting (see Results for why) ## Data **No public, row-level loan-default dataset exists for the Nigerian micro-lending / POS-agent market** that kind of individual borrower data is proprietary to lenders and protected under banking privacy regulation. This was confirmed by checking the Central Bank of Nigeria …