Integrating Cloud IoT and Federated Learning for Privacy-Preserving Autism Monitoring

Abstract

Constant observation of behavior can significantly enhance the quality of autism treatment but conventional centralized AI models are extremely dangerous in terms of privacy and sharing data. This paper suggests a federated learning (FL)-based hybrid cloud-IoT system to guarantee privacy-aware behavioral monitoring. Wearable and environmental sensor data were locally processed and global updates to models were done through an encrypted cloud aggregator. A test conducted on 48 children with autism showed that FL managed to attain the same accuracy as the centralized model with the transmission of the raw data cut by 96 percent. The findings confirm the practicability of federated IoT learning in a healthcare setting with privacy, latency, and personalization as key considerations.