Fall Risk Assessment Using Single IMU

The development of fall risk assessment is essential as falls are prevalent among older people. Despite this urgency, current assessment methods predominantly rely on subjective evaluations or simplistic statistical analyses. This study focuses on developing models capable of categorizing older participants into those with a history of falls and those without using a simple functional test, a single inertial sensor and machine learning algorithms. A total of 68 older participants
$(73.7\pm 8.9\ \mathbf{yrs})$
were recruited for the study. Participants were categorized into fallers
$(n=25,\ 78.8\pm 7.8\ \mathbf{yrs})$
and non-fallers
$(n=43,\ 70.7\pm 8.3\ \mathbf{yrs})$
. An inertial measurement unit positioned at the lower back level was employed to capture both acceleration and angular rotation data from participants during a repeated sit-to-stand test. The angular rotation data was initially used to identify the phases of sit-to-stand and stand-to-sit. Subsequently, a set of 144 time and frequency domain features was extracted from the acceleration data. Employing various data augmentation techniques, seven machine learning models were developed to differentiate older participants with a recent history of falls from those without falls over the previous two years. The significance of this study is that it showcases the use of one inertial measurement unit and a simple functional test that can be seamlessly integrated into everyday devices, allowing for remote fall risk assessment. This innovative approach demonstrates the effectiveness of the Random Forest model combined with undersampling, achieving an accuracy of 0.86. This study serves as a proof of concept for future large-scale investigations aimed at assessing the risk of falls in older populations.