Gait analysis using machine learning : An implementation of temporal convolution networks on gait events and gait phase

Abstract: This thesis aims to explore the implementation of a temporal convolution network (TCN)on gait event detection and gait phase detection. The performance of the model is eval-uated based on whether the task is gait event detection or gait phase detection. For gaitphase detection the TCN models performance is evaluated by comparing its results to theones of a simpler ML structure, namely a support vector machine (SVM).The TCN performance evaluation for gait phase detection is more nuanced as SVMmodels do not perform very well in these types of tasks. Therefor an approach where theperformance is evaluated using custom metrics and k-fold cross validation to check howconsistent the results are when training and testing data change. The effects of windowsliding and low pass filtering of data are also explored for gait event detection purposes.