Machine learning predicts farmer creditworthiness well in a small IMFI dataset (Random Forest R²=0.87; Gradient Boosting F1=0.91), and blockchain is suggested to secure Shariah-compliant records; however, excessive missing data and no out-of-sample testing weaken the case for immediate scale-up.

Main Finding

Machine learning combined with blockchain can produce an adaptive, Shariah-compliant credit-scoring system for Islamic Microfinance Institutions (IMFIs). In this study, Random Forests achieved the best continuous-score prediction (R² = 0.87, MSE = 12.4) while Gradient Boosting produced superior binary creditworthiness classification (F1 = 0.91; precision = 0.89; recall = 0.93). Blockchain was proposed to secure and make transparent borrower data in a way that preserves Islamic financial principles.

Key Points

• Problem: Conventional credit-scoring models struggle to (a) incorporate non-financial/behavioral data and (b) meet Shariah requirements (no riba, fair-dealings constraints).
• Proposed solution: An adaptive Shariah-compliant scoring framework that combines machine learning (ML) for predictive accuracy with blockchain for secure, transparent data management aligned with Islamic finance rules.
• Dataset: 1,275 farmers with 52 weeks of transaction data per farmer; original transaction data had ~53% missingness which was addressed in preprocessing.
• Preprocessing: Missing transactions handled (imputation/cleaning steps) and weekly financial activity was aggregated (summed) to construct predictors for ML models.
• Models evaluated: Linear Regression, Random Forest, Gradient Boosting.
  • Regression outcome: Random Forest best (R² = 0.87; MSE = 12.4).
  • Binary classification (creditworthy / not): Gradient Boosting best (F1 = 0.91; precision = 0.89; recall = 0.93).
• Blockchain role: Data integrity, tamper-evidence, auditable records and enhanced transparency while embedding Shariah-compliance constraints into data sharing and governance.
• Broader claims: The framework can raise predictive accuracy and operational efficiency in IMFIs, support ethical financial inclusion for marginalized communities, and align with sustainability goals.

Data & Methods

• Population/sample: 1,275 farmers; panel transaction history spanning 52 weeks per individual.
• Data quality: Substantial missingness (~53% of transaction entries) which was explicitly addressed prior to modeling.
• Feature engineering: Weekly transaction values aggregated (summed) to create time-aggregated predictors; presumably behavioral features derived from transaction patterns.
• Models compared:
  • Linear Regression (baseline for continuous score prediction)
  • Random Forest (tree ensemble; best for regression here)
  • Gradient Boosting (tree ensemble; best for classification here)
• Evaluation metrics:
  • Regression: R-squared (R²) and Mean Squared Error (MSE). Random Forest: R² = 0.87, MSE = 12.4.
  • Classification: Precision, Recall, F1. Gradient Boosting: Precision = 0.89, Recall = 0.93, F1 = 0.91.
• System design: Blockchain layer for secure, auditable storage and sharing of borrower records; ML layer for scoring and decision support; a Shariah-governance overlay to enforce compliance rules in data use and product design.

Implications for AI Economics

• Improved credit allocation: Higher predictive accuracy reduces information asymmetries and adverse selection, enabling IMFIs to extend services to previously underserved or informal borrowers while maintaining Shariah compliance.
• Cost and operational effects: Automated scoring reduces manual underwriting costs and turnaround time, potentially lowering transaction costs and enabling scale in microfinance markets.
• Financial inclusion and welfare: More accurate, Shariah-compliant scoring can increase access to finance for marginalized communities (e.g., smallholder farmers) consistent with ethical and religious constraints.
• Market structure and competition: Technology-enabled IMFIs may gain competitive advantage, encouraging product innovation (profit-and-loss sharing contracts, asset-backed financing) that fits Islamic norms.
• Transparency and trust: Blockchain-backed records increase auditability and borrower trust—important for institutions whose legitimacy relies on compliance with religious and ethical norms.
• Risks and governance needs:
  • Data quality and missingness: High missingness (53%) highlights sensitivity to imputation choices and potential bias—requires robust validation and field testing.
  • Model risks: Need for explainability, fairness audits, and Shariah oversight to ensure models do not encode discriminatory or non-compliant practices.
  • Implementation barriers: Costs, technical capacity, regulatory acceptance, scalability of blockchain, and privacy concerns must be managed.
• Policy and macro impacts: If scaled, such systems could alter credit supply to low-income and informal sectors, with implications for economic growth, income distribution, and the stability of Islamic financial markets.
• Research directions: Field trials, longitudinal impact evaluation, interoperability with regulators’ systems, cost–benefit analyses, and work on interpretable, Shariah-aware ML algorithms.