Analysis of loan default for micro finances in Zimbabwe using logistic regression and neural networks .

Author

Hove, Takudzwa R

Title

Analysis of loan default for micro finances in Zimbabwe using logistic regression and neural networks .

Abstract

Microfinance institutions (MFIs) in Zimbabwe play a critical role in promoting financial inclusion, yet they face significant threats to sustainability due to rising loan default rates. These defaults undermine institutional stability and restrict access to capital for low-income and underserved populations. This study explores the drivers of default risk and aims to improve predictive accuracy through a hybrid modeling approach that combines Logistic Regression and Feedforward Neural Networks (FFNN). Using secondary data from five selected MFIs between September 2023 and August 2024, key borrower-level, institutional, and macroeconomic variables were analyzed to identify significant predictors of default. The logistic regression model highlighted interest rates as a significant factor influencing default, whereas other variables such as age and credit score had limited statistical impact. Despite class imbalance in the dataset, logistic regression provided more interpretable and better-calibrated results compared to FFNN. However, both models struggled to detect true defaulters. The FFNN model demonstrated higher recall for non-defaulters but underperformed in identifying defaults. Model performance was evaluated using metrics such as accuracy, recall, F1 score, and AUC-ROC, with logistic regression achieving the highest AUC of 0.59. Risk segmentation based on predicted probabilities enabled grouping clients into risk bands, aiding strategic loan management. Policy recommendations include managing interest rate ceilings, improving borrower data collection, and training MFI staff in risk modeling. Future research should investigate solutions for class imbalance, integrate behavioral and digital transaction data, and test alternative machine learning models like LSTM for enhanced forecasting. This study contributes practical insights into data-driven credit risk management and supports sustainable lending practices in Zimbabwe’s microfinance sector.

Date

June 2025

Publisher

BUSE

Keywords

Logistic Regression

Machine Learning

Credit Scoring

Microfinance

Supervisor

Mr B. Kusotera