Interpretable Multimodal AI for Hormonal Risk Prediction from Biosignals and Biometric Inputs

Hormonal imbalances are often challenging to detect early due to the complex interplay of physiological signals and the absence of visible symptoms. This paper introduces an interpretable multimodal artificial intelligence (AI) framework designed to predict hormonal risk by leveraging biosignals and biometric inputs. The proposed system integrates data such as electrocardiograms (ECG), photoplethysmography (PPG), skin temperature, and anthropometric features using a fusion-based deep learning architecture. To ensure clinical relevance and transparency, the model incorporates attention-based interpretability and SHAP (SHapley Additive explanations) to highlight influential features contributing to predictions. Experimental results demonstrate high predictive performance across multiple hormonal risk categories, including thyroid and cortisol irregularities, while maintaining explainability for healthcare practitioners. This approach paves the way for non-invasive, real-time monitoring and early intervention in hormone-related health conditions.