A Novel Fusion of Explainable Sparse Evoked Tiny Machine Learning Algorithms for an Effective Detection of Autism Spectrum Disorder Using IoT-Based EEG Signals

Abstract

Autism Spectrum Disorder (ASD) is a complex neurological condition that typically shows up in early childhood. It affects how individuals interact with others and communicate. The impact can vary from person to person. Identifying it early and providing ongoing support can greatly improve outcomes. Electroencephalography (EEG) is used to study unusual brain activity and irregular neural patterns linked to ASD. Machine Learning (ML) and Deep Learning (DL) techniques have been applied to better predict and diagnose ASD, as well as support those living with it. However, these algorithms suffer from the computational overhead and non-interpretability black box problems, which limit the applicability for pervasive monitoring that aids in an early exploration of ASD patterns. To overcome this challenge, this research paper proposes a novel Internet of Things (IoT) based explainable sparse tiny ML algorithm, which has pipelined stages of pre-processing, sparse graphical feature extraction, model design and finally the deployment in resource-constrained devices. The performance of this new method is measured using accuracy, precision, recall, and F1-score, and it is compared with other existing methods. To assess how easy it is to understand, the study uses SHAP analysis. The results show that the proposed system has a high accuracy of 0.99, precision of 0.984, recall of 0.98, and F1-score of 0.99, along with efficient energy use on the hardware. It highlights the value of lightweight ML models and their ability to offer clear insights, which is important for creating better systems to understand and support individuals with ASD.