Accelerometry-Based Classification of Wheelchair Propulsion Patterns Using Machine Learning Techniques

Vishnu Ambur1, 2 ; Dan Ding PhD1 ; Asim Smailagic PhD2 ; Dan Siewiorek PhD2 ; Brian French2 , Alicia Koontz1
Human Engineering Research Laboratories, VA Pittsburgh Healthcare System, Pittsburgh, PA 152061
Institute for Complex Engineered Systems, Carnegie Mellon University, Pittsburgh, PA 152132


Previous studies classified manual wheelchair propulsion patterns based on biomechanical analysis and expert observations of two-dimensional hand movement plots. The purpose of this study was to automatically classify propulsion patterns based on wrist acceleration collected from a wearable sensor platform eWatch using machine learning techniques. A test pilot wearing the eWatch on the right wrist propelled a manual wheelchair over tile floor, low-pile carpet, high-pile carpet, asphalt, and a dynamometer with four propulsion patterns while acceleration data was collected. Two machine learning algorithms were applied and the k-nearest neighbor (kNN) algorithm yielded higher classification accuracy with an average of 85.40% (range, 60% to 100%) over the five surfaces. This study will contribute to monitor wheelchair propulsion activities in the natural environment of wheelchair users.


Wheelchair, Propulsion Pattern, Accelerometer, Machine Learning, eWatch


The work is supported by the Paralyzed Veterans of America #2486 and the National Science Foundation under Cooperative Agreement EEC-0540865 (Research Experiences for Undergraduates Supplement)


Vishnu Ambur
108 Flag Creek Road
Yorktown, VA 23693
Phone Number: 757-869-0253