Machine Learning Based Regression for Prediction of Ankle Angle Using IMU Sensor Data

Gait is known as the pattern or style of walking. It is a complex process involving musculoskeletal, nervous, and psychological factors. A good understanding of gait can be useful in various fields involving biomechanics of human movement such as design and development of human assistive device, gait analysis of post-operative hip, knee, and ankle surgery patients and analyzing abnormal gait caused due to various conditions. This study uses a machine learning approach to predict a joint angle from other two joint angles during walking from ten normal subjects. In this work, ankle angle is predicted for dorsiflexion/plantarflexion, abduction/adduction and internal/external rotation movement from hip and knee angles. This approach uses Inertial Measurement Unit (IMU) sensors to capture data of interest (joint angles) from the lower body. Several machine learning algorithms are implemented, out of which, polynomial regression gives the highest average R2 value of 0.96. The results of this work will be a good contribution to develop simplistic assistive devices with lesser number of sensors for rehabilitation for patients and assist them in their day-to-day activities and also to develop an augmentative assistive device.