Abstract. The paper analyzes the real set of large-volume medical and statistical data to be used for recognition of actions of medical workers on the basis of readings of accelerometers at a particular moment of time. During the recognition, deep belief network is applied on unlabeled data, and then trained with supervised learning by backward propagation of errors. The obtained results show a higher recognition accuracy as compared with the basic methods A significant improvement is achieved as to the duration of actions of medical staff.
Keywords: deep belief network, accelerometer, deep neural network.
Галкин Александр Анатольевич,
кандидат физ.-мат. наук, ассистент кафедры Киевского национального университета имени Тараса Шевченко,
e-mail: galkin.o.a@gmail.com