Jul 2026· Journal of Advances in Mathematics and Computer Science· Vol 41, pp. 276-308· 0 citations
Abstract
Background: Access to credit remains a major challenge for women-owned microenterprises in sub-Saharan Africa, yet existing loan default prediction models often lack transparency and interpretability. This study aimed to develop and evaluate a SHAP-based explainable machine learning framework for predicting loan default among women borrowers using Savings and Credit Cooperative Organizations (SACCO) lending data in Kenya.
Methods: We analysed 3,880 loan records from Kenyan SACCOs, comprising variables across five domains: demographic characteristics, business attributes, loan terms, group-lending mechanisms, and savings/credit history. SSix machine learning algorithms, Logistic Regression, Decision Tree, Random Forest, Gradient Boosting, XGBoost, and LightGBM, were implemented and compared. The Synthetic Minority Oversampling Technique (SMOTE) addressed class imbalance (82.5% non-default versus 17.5% default). Model performance was assessed using classification metrics, probability calibration (Brier score), computational efficiency, and 95% bootstrap confidence intervals. Model interpretability was evaluated using SHapley Additive exPlanations (SHAP), with Random Forest selected as the primary interpretable model.
Results: Ensemble learning methods outperformed traditional approaches across most performance metrics. Random Forest achieved the highest predictive performance with an F1 score of 0.903 (95% CI: 0.891–0.914) and recall of 0.999 (95% CI: 0.997–1.000), correctly identifying 99.9% of actual defaulters. Gradient boosting models (LightGBM, XGBoost, Gradient Boosting) demonstrated superior probability calibration (Brier = 0.147) and computational efficiency, requiring less than 0.6 seconds training time and producing model sizes below 0.25 MB. SHAP analysis identified loan amount, loan term, and interest rate as the dominant predictors of default, while training attendance and group enforcement mechanisms were associated with reduced default risk.
Implications for Practice: The proposed SHAP-based framework provides financial institutions with transparent, actionable insights for credit risk assessment. By identifying modifiable loan-design factors and protective behavioral mechanisms, SACCOs can enhance portfolio quality, implement risk-based lending strategies, and expand inclusive access for women entrepreneurs without sacrificing predictive accuracy.
Conclusion: Accurate and interpretable loan default prediction is feasible in microfinance settings using SHAP-based explainable machine learning. Loan characteristics are the primary drivers of default risk, while behavioral and group-based factors contribute to repayment performance. The framework supports improved lending decisions, enhanced financial inclusion, and better portfolio quality management for women-owned enterprises in developing economies.
Credit default prediction has become an important application of machine learning in the banking and financial sector, as it helps financial institutions identify potential loan defaulters and support informed lending decisions. Although machine learning models often provide high predictive performance, many of them function as black-box system, making it difficult for financial analysts and decision-makers to understand the reasoning behind their predictions. This lack of transparency can reduce user trust, particularly in high-stakes financial applications where explainable decisions are essential. To address this challenge, this study explores the use of Explainable Artificial intelligence (XAI) techniques to improve the interpretability of credit default prediction. A Random Forest classifier was developed using a publicly available credit default dataset containing financial attributes such as employment status, bank balance, annual salary, and loan default status. The dataset was preprocessed and partitioned into training and testing sets before model development. To explain the prediction process, SHAP (Shapley Additive Explanations) and LIME (Local Interpretable Model-Agnostic Explanations) were integrated with the trained Random Forest model. SHAP was used to provide both global and local explanations by identifying the overall importance and contribution of individual features, while LIME generated instance-level explanations to illustrate how specific features influenced individual predictions. The explanation results were presented through visualizations, including feature importance plots, waterfall plots, and local explanation graphs, allowing a clearer understanding of the model's decision-making process. The findings demonstrate that the combined use of SHAP and LIME enhances the transparency and interpretability of the Random Forest model by providing complementary perspectives on feature contributions. This study highlights the practical value of explainable machine learning in developing more understandable, trustworthy, and accountable credit risk assessment systems for real-world financial decision-making.
Muskan, B. Sidhu· International Journal of Com...· 0 citations
An Explainable Machine Learning (XML) framework for credit risk assessment that combines an ensemble classifier, integrating XGBoost, Random Forest, and LightGBM, with an integrated SHAP-and-LIME explainability layer is proposed and evaluated using a large-scale retail and priority-sector loan dataset drawn from public sector, private sector, regional rural, and small finance bank segments operating in India.
A. Agrawal, Vaibhav C. Gandhi· International journal of com...· 0 citations
It is argued that predictive accuracy and regulatory transparency are not competing objectives but complementary necessities for institutional survival in Nepal’s cooperative sector.
S. K. Sahani, Tsair-Fwu Lee, Digvijay Pandey et al.· Journal of Intelligent Decis...· 0 citations
Loan approval prediction is central to financial risk management, where lenders need models that are both accurate and interpretable. We compared five machine learning classifiers on a loan approval dataset: Random Forest, XGBoost, LightGBM, Logistic Regression, and Support Vector Machine. The original 45,000-sample dataset was reduced to 20,000 for training due to computational constraints. We applied SHAP TreeExplainer to interpret the best-performing model. XGBoost achieved the highest AUC (0.9747) and accuracy (0.931). SHAP identified previous loan status, personal income, loan percentage, and loan interest rate as the top four features by importance. The analysis also traces how each feature shifts individual predictions toward approval or rejection. These findings give practitioners evidence for model selection in loan approval settings and produce explanations that meet regulatory transparency requirements.
Shengze Xu· Advances in Economics, Manag...· 0 citations
Credit risk assessment is pivotal to the sustainability of micro-lending institutions, particularly in emerging economies such as Ghana, where conventional evaluation methods remain predominantly manual and subjective. Traditional approaches, which rely on face-to-face interviews, personal judgments, and simple background checks, are vulnerable to human biases, inconsistencies, and inefficiencies that contribute to elevated default rates and broader financial instability. This study investigates the application of machine learning (ML) techniques, specifically Random Forest (RF), Extra Tree Classifier (ETC), and a probability-averaged Ensemble Classifier, to enhance credit risk assessment in Ghanaian micro-lending institutions. Using a quantitative experimental research design, the study analysed 32,581 loan records drawn from Tepa Man Microfinance Institution. Data preprocessing included missing-value imputation, one-hot encoding, and class balancing via random oversampling, applied exclusively to the training set. Model performance was evaluated through 10-fold stratified cross-validation using accuracy, precision, recall, F1-score, AUC-ROC, Cohen's Kappa, and Matthews Correlation Coefficient (MCC). Hyperparameters were set to scikit-learn defaults (n_estimators = 100, random_state = 42) to ensure reproducibility. The Random Forest and Extra Tree Classifiers each achieved a mean accuracy of 99.33% and an AUC-ROC of 0.9997, results that are consistent with the high-quality, real-world dataset and are critically interpreted in the context of potential overfitting risks. Feature importance analysis identified the loan-to-income ratio and interest rate as the dominant predictors of default. The Ensemble Method, which averages class probabilities across both base models, achieved 84.25% accuracy and an AUC of 0.9231, demonstrating stronger generalization than the individual classifiers. The study concludes that integrating ML models can substantially improve the accuracy, consistency, and reliability of credit risk evaluations, thereby reducing default rates and supporting financial inclusion in Ghana's microfinance sector.
P. Addo, Samuel Kofi Akpatsa, Emmanuel Mensah et al.· Journal of Applied Social Sc...· 0 citations
Microfinance institutions (MFIs) play a critical role in improving financial inclusion in Kenya; however, high loan default rates continue to threaten their financial sustainability. This study aimed to develop and compare the performance of Logistic Regression, Naïve Bayes and Support Vector Machine (SVM) models in predicting loan default among clients of MFIs in Makueni County, Kenya. The study adopted a quantitative research design and used secondary data comprising 4,592 borrower records obtained from selected MFIs operating within the county. Data analysis was done using python programming. Borrower socio-economic and financial attributes were extracted from loan records and preprocessed through data cleaning, normalization an. encoding procedures. To ensure robust and unbiased evaluation, Stratified 5-fold cross-validation combined with Grid Search CV hyperparameter tuning was applied across all models. Model performance was assessed using accuracy, precision, recall, F1-score, specificity, and Area Under the Curve (AUC). The results showed that the SVM model achieved the highest predictive performance (accuracy = 0.857, AUC = 0.914), followed by Logistic Regression (accuracy = 0.830, AUC = 0.904), while Naïve Bayes performed least effectively (accuracy = 0.740, AUC = 0.769). The findings demonstrate that SVM provides superior classification ability in capturing complex borrower patterns, while Logistic Regression remains a strong and interpretable baseline model. The study concludes that machine-learning models, particularly SVM, significantly improve credit risk prediction in microfinance environments and can support data-driven lending decisions in rural financial institutions.
J. Musau, Divya Kasina, Dr. Ayubu Anapapa· International journal of res...· 0 citations