HealthFusion: Real-Time Human Activity Recognition and Heart Rate Estimation Using Deep Learning on Smartphone Sensors

This paper presents HealthFusion, a real-time health monitoring system integrating Human Activity Recognition (HAR) and heart rate estimation using deep learning optimized for smartphones. The HAR component uses a four-layer MLP achieving 92.59% accuracy on the UCI HAR dataset. The heart rate module achieves 2.97 BPM MAE and R² of 0.9629, outperforming commercial devices (4–7 BPM). Both models run in a React Native and FastAPI application with user authentication, dashboards, history, and SQLite storage. Comprehensive testing validates reliability for telemedicine and preventive healthcare.