ANN and SVM Classifiers in Identifying Autism Spectrum Disorder Gait Based on Three-Dimensional Ground Reaction Forces

Abstract

Autism spectrum disorder (ASD) is a complex and lifelong neurodevelopmental condition that occurs in early childhood and is associated with unusual movement and gait disturbances. An automated and accurate recognition of ASD gait provides assistance in diagnosis and clinical decision-making as well as improving targeted treatment. This paper explores the use of two well-known machine learning classifiers, artificial neural network (ANN) and support vector machine (SVM) in distinguishing ASD and normal gait patterns based on prominent gait features derived from three-dimensional (3D) ground reaction forces (GRFs) data. The 3D GRFs data of 30 children with ASD and 30 typically developing children were obtained using two force plates during self-determined speed of barefoot walking. Time-series parameterization techniques were applied to the 3D GRFs waveforms to extract the significant gait features. The stepwise method of discriminant analysis (SWDA) was employed to determine the dominant GRF gait features in order to classify ASD and typically developing groups. The 10-fold cross-validation test results indicate that the ANN model with three dominant GRF input features outperformed the kernel-based SVM models with 93.3% accuracy, 96.7% sensitivity, and 90.0% specificity. The findings of this study demonstrate the reliability of using the 3D GRF input features, in combination with SWDA feature selection and ANN classification model as an appropriate method that may be beneficial for the diagnosis of ASD gait as well as for evaluation purpose of the treatment programs.