Classification of anomalous gait using Machine Learning techniques and embedded sensors

Abstract

Human gait can be a predictive factor for detecting pathologies that affect human locomotion according to studies. In addition, it is known that a high investment is demanded in order to raise a traditional clinical infrastructure able to provide human gait examinations, making them unaffordable for economically vulnerable patients. In face of this scenario, this work proposes an accessible and modern solution composed of a wearable device, to acquire 3D-accelerometer and 3D-gyroscope measurements, and machine learning techniques to classify between distinct categories of induced gait disorders. In order to develop the proposed research, it was created a dataset with the target label being 4 distinct and balanced categories of anomalous gait. The machine learning techniques that achieved the best performances (in terms of accuracy) in this dataset were through the application of Principal Component Analysis algorithm following of a Support Vector Machines classifier (94 \%). Further, an architecture based on a Feedforward Neural Network yielded even better results (96 \%). Finally, it is also presented computational performance comparison between the models implemented.

References in corpus (4)

• Deep 1D-Convnet for accurate Parkinson disease detection and severity prediction from gait
• Sensor-based Gait Parameter Extraction with Deep Convolutional Neural Networks
• Systematic Comparison of the Influence of Different Data Preprocessing Methods on the Performance of Gait Classifications Using Machine Learning
• Feature-Set-Engineering for Detecting Freezing of Gait in Parkinson's Disease using Deep Recurrent Neural Networks