Vol. 2, Issue 1, Part A (2025)

The growing complexity of financial markets has necessitated the development of predictive models that not only achieve high accuracy but also provide transparent, interpretable insights into underlying risk dynamics. This study investigates Explainable Predictive Models for Financial Risk Forecasting (Peer-Reviewed and Revised Version), focusing on the integration of interpretability constraints within machine learning frameworks to balance predictive power with transparency. A hybrid architecture was developed and tested across three critical domains—credit default, equity volatility, and systemic risk forecasting—using financial time-series data spanning a decade. The model employed both post hoc and intrinsic explainability techniques, including SHAP, LIME, and attention-based regularization, to evaluate local and global explanation fidelity. Statistical validation through the Diebold-Mariano and DeLong tests confirmed that interpretable-regularized models maintained accuracy comparable to black-box counterparts while demonstrating superior stability and expert-aligned explanations. Key performance metrics, such as the Global Consistency Index (GCI) and SHAP Stability Index (SSI), indicated substantial improvements in explanation reliability across varying market regimes. The findings underscore that embedding interpretability during model training enhances not only transparency and auditability but also user trust and compliance readiness within financial institutions. Practical recommendations emphasize early-stage integration of explainability constraints, the establishment of interpretability evaluation metrics, and the inclusion of interdisciplinary collaboration in model design and validation. Overall, the research provides a robust framework for developing explainable, high-performing, and regulation-compliant predictive systems that can support sustainable and ethical decision-making in the modern financial sector.