FedMMIoT: A Federated Learning Framework for Detecting Autism Spectrum Disorders Using Multimodal Internet of Things

This paper explores the application of federated learning in distributed Autism Spectrum Disorder (ASD) detection using Internet-of-Medical-Things devices (IoMT). The proposed method addresses some of the challenges in processing IoMT data, such as dealing with different types of data from multiple sensors and scalability to accommodate multiple edge devices. This system maintains confidentiality while improving the accuracy of classifications using collaborative learning techniques among multiple edge devices. Federated averaging makes model aggregation possible and thus makes collaborative learning possible among multiple devices and not at a central location to avoid confidentiality breaches. From experiments, there is promising accuracy of 87% and a loss of 0.39 using federated learning. Additionally, this proposed system reduces latency values up to 50%, indicating high efficiency in real-time ASD detection systems. The proposed system is more energy-efficient compared to other IoMT designs in edge computing and cloud computing systems. Edge computing coupled with caching makes this proposed system more efficient in terms of energy and has high adaptability and efficiency in IoMT using multiple devices for ASD detection.