Machine learning model for predicting shear forces at the body-seat interface in wheelchair users: A novel approach

Abstract

Among the mechanical factors contributing to pressure injuries, shear forces at the body-seat interface play a critical role. This study introduces a novel machine learning approach to predict these forces, using data from pressure mapping systems and a multi-adjustable experimental seat. A supervised learning model was trained on measurements collected from individuals without disabilities and later evaluated on both this group and a cohort of wheelchair users. The selected model - a Random Forest Regression - relied on six input features: a calculated variable, backrest force, feet normal force, seat pan force, backrest area, and the location of the backrest center of pressure. It demonstrated promising accuracy, with an average error below 20% for individuals without disabilities and for wheelchair users whose shear forces were within a similar range. However, performance declined for wheelchair users exhibiting significantly lower shear forces. To improve generalizability, future work will expand the dataset to include participants with more diverse anthropometric characteristics and a broader range of seated postures.

Keywords: Machine learning; prediction; pressure injury; shear force; wheelchair user.