Federated-Learning-Based Smart Healthcare Framework for Yellow Fever Detection

Deep learning (DL) has proven to be an efficient methodology for yellow fever determination, yet privacy concerns from restricted data sharing by medical institutions compromise its performance. While existing literature has explored federated learning (FL) for privacy-preserving model training, this research introduces a novel FL-inspired methodology that integrates multimodal data sources of audio and visual data to identify yellow fever symptoms. This approach employs the Dynamic Integration Approach to estimate health severity, marking a significant advancement over traditional methods that typically rely on single data modalities. Comparative assessments with state-of-the-art methods demonstrate superior performance, with statistical measures showing Accuracy (98.57%), Precision (99.85%), F-Score (96.53%), and Recall (94.72%). Additionally, improved results in delay latency (11.80 s) and stability (71%) highlight the approach’s efficiency. This work has broader implications for healthcare and public health systems, offering a scalable and privacy-preserving solution for epidemic monitoring.