Identifying Autistic Children by Machine Learning Models with Features Extracted from Wearable Inertial Sensors

Since children with autism spectrum disorder (ASD) exhibit impaired motor functioning and atypical motor patterns, this study aimed to identify ASD using low-cost wearable inertial sensors through the Internet of Thing (IoT) technique. Three rotation metrics (i.e., total rotation time, rotation amplitude, asymmetry) were extracted using low-cost wearable inertial sensors during a wrist rotation imitation task (completed within 1 minute). Five machine learning (ML) based ASD classifiers, using k-nearest neighbors (KNN), linear discriminant analysis (LDA), naive Bayes (NB), support vector machines (SVM), and random forests (RF), were constructed, in which the suggested three metrics were considered as input variables. Findings showed that the ML-based classifier with the naive Bayes algorithm outperformed than other four algorithms, and achieved a maximal classification accuracy of 88% and a maximal AUC value of 0.91. In conclusion, ML-based classifiers with rotation features may be applied for a rapid screening of ASD.