Mutethia-ds/SME-Credit-Scoring-System

SME credit scoring using alternative data for loan default prediction in Kenya # SME CREDIT SCORING FOR LOAN DEFAULT PREDICTION IN KENYA ## Project Overview Access to credit remains a major challenge for Small and Medium Enterprises (SMEs) in Kenya. Traditional credit scoring methods often rely on formal credit histories and financial records, excluding many SMEs that operate in informal or semi-formal environments. This project develops an alternative credit scoring system that leverages digital financial behavior, business activity indicators, utility payment patterns, and mobile money transaction data to predict loan default risk. Multiple machine learning models were evaluated, compared, and combined to identify the most effective approach for SME credit risk assessment. The resulting system generates default probabilities, credit scores, risk categories, and lending recommendations that can support financial institutions, fintech companies, SACCOs, and digital lenders in making data-driven credit decisions. --- ## Research Title **An Ensemble Machine Learning Approach to SME Credit Scoring Using Alternative Data for Loan Default Prediction in Kenya** --- ## Problem Statement Many Kenyan SMEs lack sufficient credit history to qualify for loans under traditional credit assessment frameworks. As a result: * Creditworthy businesses may be denied financing. * Financial institutions face increased uncertainty when lending. * Informal and underserved businesses remain financially excluded. This project investigates whether alternative data sources can be used to build an accurate and interpretable machine learning-based credit scoring system. --- ## Objectives ### Main Objective To develop an ensemble machine learning credit scoring model that predicts SME loan default risk using alternative data sources. ### Specific Objectives * Develop predictive models for SME loan default. * Engineer meaningful financial and behavioral features from alternative data. * Compare the performance of multiple machine learning algorithms. * Gene …