Reach and throw movement analysis with support vector machines in early diagnosis of autism

Abstract

Movement disturbances play an intrinsic part in autism. Upper limb movements like reach-and-throw seem to be helpful in early identification of children affected by autism. Nevertheless few works investigate the application of classifying methods to upper limb movements. In this study we used a machine learning approach Support Vector Machine (SVM) for identifying peculiar features in reach-and-throw movements. 10 pre-scholar age children with autism and 10 control subjects performing the same exercises were analyzed. The SVM algorithm proved to be able to separate the two groups: accuracy of 100% was achieved with a soft margin algorithm, and accuracy of 92.5% with a more conservative one. These results were obtained with a radial basis function kernel, suggesting that a non-linear analysis is possibly required.