AI FOR MELTDOWN DETECTION IN AUTISM USING WEARABLE SENSORS.

Abstract

Autism spectrum disorder is a neurodevelopmental disorder that is associated with many symptoms, such as impairments in social skills, communication, and abnormal behaviors. Children on the spectrum exhibit atypical, restricted, repetitive, and challenging behaviours. The occurrence of such behaviours poses challenges to caregivers and therapists during therapy sessions. In this study, we investigate the feasibility of integrating wearable sensors and machine learning techniques to detect the occurrence of challenging behaviours among children with autism in real-time. Children wore a wearable device, which collected physiological data in five sessions. The video recordings of the sessions were analyzed to identify the instances of challenging behaviours. Four machine learning techniques were used to leverage various features extracted from the wearable sensors to automatically detect challenging behaviors. The best prediction performance was observed when the XGBoost algorithm was used with all gathered features (i.e., accuracy of 99%). Physiological features were found to be more effective than kinetic ones for the prediction task. Among various physiological features, the heart rate was the main contributing feature in the detection of challenging behaviours. Furthermore, experiments revealed that changes in the HRV parameter (i.e., RMSSD) correlated to the instances of challenging behaviours. The findings of this work motivate research towards methods of early detection of challenging behaviours which enable timely intervention by caregivers and parents.