A study on auto-scoring algorithm for balance assessment

초록

In this paper, we studied scoring algorithm for BBS (berg balance scale) assessment using IMU (inertial measurement unit) and machine learning methods. Thirty-five patients with brain disease, aged 50 to 80 years, participated in the experiment. The IMU sensors were placed on their forehead, back, left wrist, right wrist, left ankle, and right ankle. The data of 3-axis acceleration roll, yaw, and pitch from each IMU were measured and the sampling rate of all signal was 100Hz. SVM (Support vector machine) and MLP (multi-layer perceptron) were compared as machine learning techniques. According to each motion of BBS task, different data from 6 IMUs was selected. The average accuracy of auto scoring of SVM and MLP models was 92% and 88%, respectively.