Comparison of Artificial Neural Networks with other Machine Learning Methods in Foot Movement Classification

Abstract

Modern prostheses can be controlled by using gait analysis data from Inertial Measurement Units compared to traditional prostheses. This article aims to classify foot movements for the robotic ankle system in lower limb prostheses to recognize motion intent and adapt to abnormal walking conditions. The statistical features are extracted from IMU data from 11 volunteers aged 20-34 and then the features are classified using machine learning. In this study, the classification accuracies of Naïve Bayes Classifier, Linear Discriminant Analysis, K-Nearest Neighbour Classifier and Support Vector Machines and Artificial Neural Networks in classifying foot movements are examined separately for the raw data and the processed data such as Euler angles and quaternions which estimate with Madwick Filter. Gait analysis data were obtained by using the Inemo inertial module LSM9DS1 work on an NRF52 including 9 DOF, triaxial gyroscope, triaxial accelerometer, and triaxial magnetometer in the Biomechanics Laboratory of the Department of Mechanical Engineering, Middle East Technical University from eleven subjects and achieved an highest classification accuracy rate of 90.9% on test data, 97.3% for training data.