Examining the Research Gap in Explainability Methods for Credit Scoring Models in Nigerian Microfinance

Explainable AI is a critical requirement for responsible credit decision-making, yet the specific explainability challenges and available methods for credit scoring models in the Nigerian microfinance context have received minimal research attention. This study examined the research gap in XAI for Nigerian microfinance credit scoring through systematic review and original empirical analysis. A systematic scoping review identified 91 publications from 2019 to 2024 on XAI in credit scoring, of which only 4 addressed sub-Saharan African microfinance contexts and zero specifically addressed Nigeria. Original empirical analysis compared LIME, SHAP, and Anchors explanation methods applied to gradient boosting credit scoring models trained on 6,800 microfinance loan records from three institutions in Onitsha and Aba. Model performance was equivalent across all three XAI-augmented configurations (AUC approximately 0.82), confirming that explainability integration did not reduce predictive accuracy. SHAP provided the most stable and consistent feature attribution, with previous default history and loan-to-income ratio identified as dominant decision drivers. LIME explanations showed higher instability across repeated queries, raising concerns for regulatory audit purposes. Borrower comprehension of model explanations was assessed through interviews with 30 loan officers and 20 borrowers, confirming that simplified natural language summaries were significantly better understood than numerical feature importance scores. The study fills an important XAI gap in Nigerian microfinance and recommends SHAP-based explanation as the standard approach, with natural language explanation summaries adopted for regulatory reporting and loan officer communication.

Keywords: explainable AI, credit scoring, microfinance, Nigeria, SHAP