Innovative healthcare systems rely on IoT devices to continuously monitor patient health and collect sensitive data. Traditional cloud-based machine learning methods raise privacy concerns and require high bandwidth. To overcome these issues, this paper proposes a lightweight Federated Learning (FL) approach that enables IoT devices to collaboratively train models without sharing raw data. The proposed system is optimized for low-power devices using model compression techniques and efficient communication strategies. Experiments on healthcare datasets demonstrate that the approach achieves high accuracy while utilizing reduced resources. This work highlights the potential of privacy-preserving, intelligent healthcare systems using FL and IoT integration.