Machine learning-based fall characteristics monitoring system for strategic plan of falls prevention

The occurrence of fall increases with age and level of physical frailty. Due to the fact that falls are unexpected and inevitable events, the fall characteristics recording to collect the related circumstance and characteristics of occurred fall events are important for strategic plan of falls prevention. However, typical clinical recording approaches suffers issues in objective and continuous assessment. In this study, a fall characteristic monitoring system is proposed to support the clinical professionals to assess the causes of fall events for fall prevention strategies, which consists of high accuracy fall event detection algorithm and fall direction identification. Eight males are recruited in this experiment and asked to perform the seven types of fall and six types of ADL. A waist-based tri-axial accelerometer is used to measure motion acceleration with 128 Hz sampling rate. The sensitivity, specificity, precision, negative predictive value, and accuracy using the hierarchical fall event detection algorithm are 99.83%, 98.44%, 98.67%, 98.44, and 99.19%, respectively. Furthermore, the overall average performances of the sensitivity, precision, and accuracy using the fall direction identification are 98.52%, 97.49%, and 97.34%, respectively, the results demonstrated that the proposed approach is fulfilled the requirements of fall characteristics monitoring system.