2017 年 83 巻 856 号 p. 17-00210
Falling from the bed is a common type of accident and places considerable burdens on patients and nurses. Structural and risk factors for the occurrence of falls have been identified, but fall prevention remains extremely difficult due to the patient’s physical, mental, and social factors and treatment environment. Most fall prevention measures involve ascertaining the risk of falls through the use of risk assessment score sheets and bed sensors, but there are few measures for active fall prediction. To develop a method for fall prediction, we applied area trajectory analysis and spectrum analysis to the characteristics of center-of-gravity variation in certain movements. We used these analysis methods and applied Support Vector Machine (SVM) that is one of the methods of machine learning. Experiments were performed with 5 healthy male and female. Each participant performed 3 movements, Reach out, Bed rail and Active, on a bed for 1 min each, during which time-series data on center-of-gravity variation were recorded. In the micro-average about unknown data, the Precision rate was 90.6%. To evaluate the movements respectively, Active were both higher in Precision rate and Recall rate. However in the Reach out has low Precision rate and that likely cause misinformation, in the Bed rail has low Recall rate and that likely cause overlook. The results of this study suggest the possibility of fall prediction through center-of-gravity analysis. In the next step about this study, need to explore the discriminate about static posture and improvement in accuracy by increasing the learning data.