This study introduces an adaptive edge-federated AI framework for real-time, privacy-preserving menstrual health prediction using contactless biosensing. Radar, PPG, and LiDAR monitor physiological signals, analyzed locally via edge learning, while federated optimization ensures secure, decentralized model training. The system dynamically adapts to individual variations, enhancing prediction accuracy, reducing latency, and preserving user privacy, providing a non-invasive next-generation menstrual health management solution.