The prediction of patient actions is a valuable mechanism for monitoring health records by tracking movements and updating the patient’s condition according to their specific problems. The healthcare system faces significant challenges, particularly in monitoring and treating children, the elderly, and those with suspected conditions. These challenges include incomplete data, a lack of real-time information, and difficulty in recognizing diverse patient actions. To address these issues, the proposed Federated Learning with Weighted Averaging-enabled Hybrid Deep Learning (FL-WA-HDL) model improves patient health monitoring by enhancing prediction accuracy. This model outperforms the traditional approaches, offering a flexible, efficient, and reliable way to predict treatment-related risks and identify resource allocation needs for better patient care. The FL-WA-HDL model is integrated with Deep Learning models, evaluated with real-time data, and yields a high accuracy of 96.86%, precision of 97.40%, and recall of 96.32% at a training percentage of 90%. These measures indicate the FL-WA-HDL model’s great potential to revolutionize the area of remote healthcare monitoring, ensuring enhanced patient outcomes through reliable and timely predictions.