Prediction of Bed-leaving Behaviors Using Accelerometer-embedded Pillow Based on Machine Learning
Hirokazu Madokoro, Nobuhiro Shimoi, Kazuhito Sato · Transactions of the Society of Instrument and Control Engineers · 2013
This paper presents a sensor system to predict behavior patterns that occur when patients leave their beds. We installed a triaxial accelerometer into a pillow of its bottom side. The features of our sensor are simple, low cost, and no restraint of patients. Moreover, we develop a recognition method of behavior patterns using unsupervised machine learning algorithms . We evaluated our method obtained using our sensor system for three subject at an environment that represents a clinical site. The recognition accuracy of the terminal sitting, which is defined a subject trying to leave the bed, is 94.4%. However, the recognition accuracy for seven behavior patterns is 56.3%. Misrecognition was remained inside sleeping, sitting and leaving in each category. The recognition accuracy is improved to 94.4% when we evaluated our method for these three categories.