Faculty Mentor(s)

Dr. Omofolakunmi Olagbemi, Computer Science; Dr. Brooke Odle, Engineering

Document Type


Event Date



A 2016 survey indicated that 39% of registered nursing respondents had reported musculoskeletal injuries after two years of regularly performing patient-handling tasks. Optical marker systems (considered the gold standard) can be used in laboratory settings to explore mechanisms of injury during patient-handling tasks, but deploying inertial measuring units (IMUs) in biomechanics allows data collection in both laboratory and clinical environments. IMU-based capture systems are also preferable to optical marker systems because they avoid marker occlusion during more complicated patient-handling tasks. The aims of our study are (1) to identify machine learning models that can accurately predict patient-handling tasks performed and the quality of posture adopted by participants performing those tasks (using data from wearable sensors — IMUs — and force plates), and (2) to determine an optimal combination of those IMUs. Using trunk-and-pelvis IMU data from two participants performing three tasks with good and poor postures, the MiniRocket machine learning model was the fastest of five models utilized (123s) and also the most accurate (98.1%). Future work includes involving additional participants and expanding the range of tasks.


This research was supported by the Howard R. and Margaret E. Sluyter Faculty Development Fund, the Clare Boothe Luce Research Scholars Program, and the Hope College Department of Computer Science.