<![CDATA[In financial data analysis, credit risk assessment has traditionally depended on financial ratios. However, a lot of useful information is hidden in company reports, especially in the textual sections. Most of the financial analytics models do not utilize the textual features of financial data. To address this gap, this work proposes the Cognitive Analytics, Multimodal and Explainability (CME-Insights) framework, which is designed to provide clear and data-driven support for financial decision making. The proposed approach combines numerical financial indicators and textual information from the Management Discussion and Analysis (MD&A) sections. These two modalities of data are extracted from SEC 10-K and 10-Q reports. The structured financial modality is drawn from a publicly available corporate credit rating dataset comprising 2,029 records. The model uses 25 financial features along with 11 text-based features. These text features capture characteristics such as sentiment, uncertainty, risk, legal tone, and readability. Loughran-McDonald financial dictionary and the VADER sentiment model are used to extract these features. The best results came from a stacked model that combines CatBoost, LightGBM, and Random Forest base learners with a Logistic Regression meta-classifier. This model achieved an accuracy of 89.36% when ticker symbols were included, and 86.24% without them. SHAP and LIME explainability approaches indicated that the features such as debt ratio and operating cash flow per share play a key role in predicting credit risk. Thus, the CME-Insights framework delivers more accurate and explainable credit risk assessment than traditional financial-ratio-only approaches.]]>
Copyrights © 2026
