Decision Tree Machine Learning to Determine Direction of Steering Force for Hospital Bed Push Handle System

Abstract

This work demonstrates the use of Decision Tree and Random Forest machine learning to determine the direction of desired movement from the forces applied to two push-handles on a novel patient transfer system the Able Innovations ALTA™ patient transfer system provides a replacement transfer method for residents in care that cannot transfer on their own that have limited mobility. This new system is significantly heavier that a hospital gurney because of the weight of the mechatronics and healthcare professionals will need power assist to transport and position the system. In this work, two loadcell sensor-based prototypes have been used to measure the input forces and direction applied by a user to two handles. These two prototypes have been placed on a hospital bed to simulate the two handles on the right and left provided to hospital staff to push the system. Machine learning is used to analyze sensor measurements from each handle to predict the movement intent. The results are presented for test pushes in 6 directions that include forward and reverse in each of straight, left, and right turns. Two models of Machine Learning (decision tree classifier and random forest classifier) have been used for 8 and 14 features and are shown to be able to predict the direction of push with high accuracy. The accuracy for 8 features using decision tree and random forest has been measured 93.5% and 97.0% respectively.