New Techniques for Real-Time Fall Prediction and Development of an Injury Prevention System
Emily Kamienski · 2023
Falls may cause serious injuries to older adults, leading to a deteriorated quality of life. Realtime fall injury prevention devices are lacking, especially for the balance impaired population who rely on mobility aids. A key challenge to the reliability of such devices is real-time fall prediction, which is a timecritical decision making process. A fall must be predicted preemptively so that the system has sufficient time to deploy the injury prevention mechanism. Furthermore, the fall predictor must have a sufficiently low false positive rate for practical use. This paper presents three techniques for improved fall prediction and a fall injury prevention device for older adults with declining balance. We first collected a novel fall dataset from human subjects undergoing diverse loss-of-balance situations while using a mobility aid. Data obtained from these human subject tests contain diverse patterns of fall cases, yet both false negative and false positive rates must be very low; otherwise, the system cannot be adopted reliably. The first modeling technique uses multiple Long-Short Term Memory networks in parallel, each tuned to an individual pattern of fall data, yielding a higher sensitivity than a baseline model trained on all the fall data. Second, to reduce the false positive rate, another Long-Short Term Memory network is constructed to predict the time remaining before the fall prevention mechanism must be activated, so that if time allows, prediction may be delayed and additional data collected. Third, confounding cases are further analyzed using a metric of data deficiency, called the Lipschitz quotient. Additional data features that lower the Lipschitz quotients, increasing data predictability, are sought and incorporated into the original signals. Separately learning individual fall patterns, and delaying fall predictions using the time remaining prediction successfully increased the number of identified falls and lowered the false positive rate. Augmenting the dataset further improved performance, and the best model had a 97% identified falls rate at a 0.17% false positive rate. An LSTM based predictive model is implemented on a novel walker-type fall prediction and prevention prototype. The walker has a small footprint for improved maneuverability, and becomes untippable when it's expandable legs are deployed in the event of a predicted fall. Thus, the older adult tethered to the untippable walker is protected from a fall. This promises immense benefits for future research on improving older adult wellbeing through real-time fall protection.