Securing Fall Detection with Federated Learning: A Privacy-Centric Approach for IoT Healthcare

Efficient and automated fall detection is crucial in healthcare, especially for monitoring elderly individuals. Rapid and accurate systems can reduce response times, mitigating the risk of prolonged injuries or fatalities. However, ensuring the privacy of sensitive data remains a significant challenge. This paper presents a privacy-preserving fall detection framework that leverages federated learning and differential privacy to collaboratively train models while keeping sensitive data on-device. To assess its effectiveness, we evaluated the framework using four widely-used public datasets: MobiAct, UNIMIB, SisFAll, and FALLALLD. The results demonstrate its robust performance and adaptability across various scenarios. A comparative analysis with existing models highlights the framework's strong results, scalability, and suitability as a privacy-focused solution for real-world IoT healthcare applications.