A machine learning-based monitoring system for attention and stress detection for children with Autism Spectrum Disorders

Abstract

The majority of children with Autism Spectrum Disorders (ASD) have faced difficulties in sensory processing, which affect their ability of effective attention and stress management. Children with ASD also have unique patterns of sensory processing when responding to the stimuli in the environment. In this study, a real-time monitoring system has been designed and developed for attention and stress detection. Comprehensive sensory information, including environmental, physiological, and sensory profile data can be collected by the system using sensors, smart devices, and a standard sensory profiling questionnaire. Data acquisition with 35 ASD children using the system prototype was successfully conducted. With the acquired data set, different machine learning models were trained to predict attentional and stress level. Among all the investigated models, Gradient Boosting Decision Tree and Random Forest obtained the best prediction accuracies of 86.67% and 99.05% on attention and stress detection respectively. The two models were then implemented into the system for automatic detection. Future work could be focusing on exploring more supportive features to improve the prediction accuracy for attention detection. Such an easily-accessed monitoring system tailored for children with ASD could be widely-used in daily life to assist ASD users with their attention and stress management.