Recognition of gait patterns in human motor disorders using a machine learning approach
João Miguel Santos Barbosa · Portuguese National Funding Agency for Science, Research and Technology (RCAAP Project by FCT) · 2019
With advanced age, the occurrence of motor disturbances becomes more prevalent and can lead to gait pathologies, increasing the risk of falls. Currently, there are many available gait monitoring systems that can aid in gait disorder diagnosis by extracting relevant data from a subject’s gait. This increases the amount of data to be processed in working time. To accelerate this process and provide an objective tool for a systematic clinical diagnosis support, Machine Learning methods are a powerful addition capable of processing great amounts of data and uncover non-linear relationships in data. The purpose of this dissertation is the development of a gait pattern recognition system based on a Machine Learning approach for the support of clinical diagnosis of post-stroke gait. This includes the development of a data estimation tool capable of computing several features from inertial sensors. Four different neural networks were be added to the classification tool: Feed-Forward (FFNN), convolutional (CNN) and two recurrent neural networks (LSTM and CLSTM). The performance of all classification models was analyzed and compared in order to select the most effective method of gait analysis. The performance metric used is Matthew’s Correlation Coefficient. The classifiers that exhibit the best performance where Support Vector Machines (SVM), k-Nearest Neighbors (KNN), CNN, LSTM and CLSTM, with a Matthew’s correlation coeficient of 1 in the test set. Despite the first two classifiers reaching the same performance of the three neural networks, the later reached this performance systematically and without the need of explicit dimensionality reduction methods.