Abstract:
Falls lead to severe public health problems resulting in devastating psychological and physical consequences for older people and their families. Therefore, fall risk assessment has been a popular eld of research in the last decades to understand the underlying reasons of the fall and eventually to identify people at high fall risk so that e ective preventive strategies for falls can be taken at an early stage. Clinical and functional fall risk assessment tools are not objective and reliable for the assessment of fall risk as they require human observation and judgement. Related studies aim to assess the fall risk based on gait parameters in a more objective way, however, they either propose obtrusive systems composed of numerous sensors or do not utilize machine learning methodology. In this study, we employ machine learning techniques and aim to develop an accurate, unobtrusive, objective and continuous fall risk assessment system based on gait parameters. For this purpose, we studied two gait analysis techniques: a Kinect-based gait analysis system, and a foot-mounted inertial sensor-based gait analysis system developed by our research group. Using the latter, experiments are conducted to extract gait parameters of 37 subjects of whom 21 have neurological conditions with gait implications. Gait features are computed from these gait parameters. Feature selection techniques are used to determine the most signi cant gait features. Based on these features, di erent machine learning algorithms are employed to classify people into high and low fall risk groups. This system enables us to identify people who are likely to experience a fall in the near future and inform their caregivers to take preventive interventions. The predictions are evaluated based on the accuracy, sensitivity, speci city and F-measure.
