THE USE OF MOTION SENSORS AND SUPPORT VECTOR MACHINE FOR CLASSIFYING SIMULATED ANKLE SPRAIN AND NORMAL MOTIONS

Abstract

Ankle sprain is one of the most common sports injuries. Our research team has developed an intelligent system to prevent the injury, and the system relies on a method to identify an ankle sprain motion. The purpose of this study is to increase the accuracy of Support Vector Machine (SVM) in classifying ankle sprain from normal motions and investigate the feasibility to employ SVM in the intelligent system. Fourteen subjects performed trials of (a) walking, (b) vertical jump, (c) stepping down a stair, and (d) jumping off a stair. Data from a motion sensor at the posterior calcaneus were used and trimmed to 230 (0.4s) and 60 (0.12s) window size, and were transformed from time to frequency domain by discrete Fourier Transform. Motion data from eleven subjects (11 out of 14) were used for training the SVM. A Radial Basis Function kernel function was employed in the SVM. Accuracy was tested on the data from another three subjects, which reached 96.1% and 93.1% for window size 230 and 60 respectively.