Machine learning based method for huntington’s disease gait pattern recognition

Abstract

Huntington’s disease (HD) is an inherited neurodegenerative disorder causing problems with mobility, cognition and mood. Gait abnormality is a potential diagnostic sign as it can occur even in the early stages of HD. We developed a machine learning method for detecting HD with gait dynamics as the model features. Concretely, standard deviation (SD) and interquartile range (IQR) were calculated for 6 gait time series sequences as 12 candidate features. An exhaustive feature and hyperparameter selector was then applied to optimize the features and hyperparameter subsets for 5 different machine learning models. Classification outcomes were determined by nested leave-one-out cross-validation (nested LOOCV) method. Support Vector Machines (SVM) achieved the highest accuracy (97.14%) without overfitting bias assumptions. Our result showed that the machine learning based method with gait dynamics features can be a complementary tool for HD diagnosis.