<![CDATA[This research presents an explainable, calibrated multimodal framework for creditworthiness scoring. It integrates financial ratios with narrative risk disclosures by using the 823 firm SEC (Securities and Exchange Commission,) filings. In this approach we have used a class-weighted, L2-regularized logistic regression within a unified pipeline. It produces calibrated probabilities via isotonic regression. We have assessed the performance across discrimination (ROC-AUC, PR-AUC), reliability (Brier, calibration curves), operating thresholds (F1, asymmetric costs), and business diagnostics. We have also included the Decision Curve Analysis to cover the pattern of net benefit across practical cutoffs. Our implementation covers an ablation study to verify multimodal gains over financial-only and text-only baselines. The present research findings state that the LightGBM + SHAP benchmark depicts better interpretability along with improved prediction accuracy. Thus. the study provide an useful insights towards effective governance and transparent system requirements which makes this suitable for effective financial risk management.]]>
Copyrights © 2026
