Classification of Lower Limb Amputees Gait Using Machine Learning Algorithms

Abstract

Human gait data, as well as other biological signals, follow distinguishable and measurable patterns that are important for the evaluation and analysis of movement. In this paper, we used spatiotemporal gait data for classification of lower limb amputee groups (transtibial and transfemoral) and individuals without amputation (control). The classification was made using algorithms based on machine learning, KNN (K-nearest neighbors) and RF (random forest). Three treadmill walking conditions were analyzed: horizontal (0º), uphill (+8%) and downhill (–8%). These conditions were important to establish in which scenarios the data are more discriminating. The classification of the data with all conditions together provided an accuracy of 75.8% and 77.7% for KNN and RF, respectively. The best classification result was obtained using the RF algorithm with the data in the downhill condition, indicating that this condition imposes greater motor demands to the participants, showing greater differences between the extracted variables.