Autism Spectrum Disorder Prediction Using M-chat Screening and EEG Signal Analysis Via Machine Learning

Abstract

The Autism Spectrum Disorder (ASD) is a developmental condition, characterized by impairments in  social interaction, communication and behavior. They have to be identified early and accurately to make a difference and intervene in time. In this paper, a web-based predictive algorithm on ASD is described, which involves a combination of the neurophysiological and behavioral data. Three key components of this system are, protected access to logins to extract data by medical workers, a behavioral screening instrument, known as the Modified Checklist of Autism in Toddlers (M-CHAT) and an EEG- Soapstone analysis module that is machine learned by the Random Forest model. This is undertaken in to steps of the workflow whereby the first part involves the application of M-CHAT to mitigate risks. EEG is recorded and pre-processed when a patient was found to be of medium or high risk. SelectKBest classifier with mutual information classify is then used to consider the most suitable features with the hopes of achieving the most powerful features with which to classify using the SelectKBest classifier. The random forest gave an accuracy of 99.53, a precision of 99.54, a recall of 99.51, a F1 -score of 99.52, and a ROC-AUC of 0.94. The findings are confirmatory of M-CHAT behavioral screening combined with further examination of the EEG signal records to forecast early ASD accurately. The proposed system provides subjective and objective diagnostic data to the clinicians. Some methods to enhance it include the implementation of real-time EEG monitoring and extending its use in clinics in future.