Assessment of heart rate variability response in children with autism spectrum disorder using machine learning

Abstract

Autism spectrum disorder (ASD) is a developmental disability that involves persistent challenges in social interaction, communication and behaviour. The purpose of this study is to apply a machine learning approach to differentiate between autistic and normal children and to evaluate the performance of different classifiers in the detection of autism disorder. Heart Rate Variability (HRV) analysis is one of the strategies used for ASD detection by assessing the autonomic nervous system (ANS), which serves as a biomarker for the autism phenotype. HRV can be derived from the photoplethysmogram (PPG). Logistic Regression, Linear Discriminant Analysis and a Cubic Support Vector Machine (SVM) were chosen to evaluate the performance of HRV features in differentiating between normal and autistic children. Three different combinations of features were selected out of 19 features in total. From the results, Logistic Regression was the best classifier to differentiate between autistic and normal children in a colour stimulus test with 100% accuracy, while Linear Discriminant Analysis was best suited in the baseline test with 90% accuracy. In conclusion, the machine learning approach could be an alternative method of making an early diagnosis of ASD in the near future.