Transforming financial documents into credit decisions using explainable artificial intelligence and optical character recognition

The increasing use of digital technology in banking and financial services has accelerated the adoption of Artificial Intelligence and Machine Learning in lending and credit scoring. While advanced ensemble models offer higher predictive accuracy than traditional statistical approaches, they are often difficult to interpret. This lack of transparency raises concerns related to fairness, regulatory compliance, and borrower trust. This study proposes an integrated Explainable Artificial Intelligence framework for automated credit scoring and lending decisions. The framework utilizes Optical Character Recognition through EasyOCR, PyPDF2, and PdfPlumber to extract financial and identity data from applicant documents, followed by data cleaning, normalization, and structured feature engineering. SHAP and LIME are applied to generate global and local explanations, enhancing interpretability. Experimental evaluation shows that XGBoost achieves 92.5% accuracy, outperforming LightGBM (90.8%), Random Forest (87%), Logistic Regression (91%), and LSTM (81%).•The methodology integrates document-based data extraction with supervised learning models to generate accurate credit scores and lending decisions.•Performance metrics including accuracy, precision, recall, F1-score, MAE, RMSE, and R² confirm the superior performance of XGBoost in credit risk prediction.•The results highlight that combining structured financial features with explainable models improves transparency, supports regulatory compliance, and enhances decision-making reliability.