An early detection of autism spectrum disorder using machine learning

Shaba Irram, Mohammad Suaib,
An early detection of autism spectrum disorder using machine learning,
Computers in Biology and Medicine,
Volume 198, Part B,
2025,
111238,
ISSN 0010-4825,
https://doi.org/10.1016/j.compbiomed.2025.111238.
(https://www.sciencedirect.com/science/article/pii/S0010482525015914)
Abstract: Background
Children with autism spectrum disorder (ASD), a developmental disease, exhibit limited and repetitive activities in addition to challenges with communication and social interaction. Numerous neurological disorders have been identified with the aid of the electroencephalography (EEG) technology, which measures the electrical activity of the brain.
Objective
The primary purpose of this study is to use EEG data to detect ASD in newborns.
Methods
Power values from baseline EEG recordings of babies are processed and analyzed to extract relevant information. Using machine learning approaches like decision tree (DT), random forest (RF), and support vector machine (SVM) and Explainable AI(SHAP), the model is trained using extracted data to differentiate between those with ASD and those who are usually developing. SHAP (SHapley Additive exPlanations) is a technique that can be used to describe the output of machine learning models. Understanding the rationale behind the predictions produced by the best-performing model is made easier with SHAP.
Results
Our machine learning model with SVM classifier (AUC = 93 %) and RF classifier (AUC = 90 %) has demonstrated exceptional performance in diagnosing infants with ASD. Novelty: Previously little-focused, this work provides a machine learning model with the use of explainable AI to identify autism spectrum disorder in children under the age of one. Early identification of ASD is challenging since children under the age of 18 months do not exhibit many behavioral indicators. Therefore, medical reports of infants' EEGs are useful in determining whether or not an infant has ASD. Applications: As a result, this model may be used to automatically diagnose ASD using the infant's EEG data.
Keywords: Electroencephalography; Autism spectrum disorder; Machine learning; Feature extraction; Neurological disorder; Autism; Early diagnosis