Gradient Boosted Decision Tree based Classification for Recognizing Human Behavior

Abstract

Human behavior prediction became an active research topic to determine the criminal and suspicious activities of a person. Gerontology deals with the everyday life activities of an individual including walking, climbing, eating, drinking, sitting and so on. It helps in ambient assisted living for the old persons in a self-reliant manner. The emergence of sensors and smart environment made the sensing process in an easier way. In general, the sensed dare classified using decision tree logic-based approach. The classification accuracy is low in case of decision tree approach. Hence, in this paper the gradient boosted tree is integrated with the decision tree approach to achieve greater accuracy. The triaccelerometer wearable sensor is used to collect the three-dimensional data of each activity of human being. The results showed that the integrated approach showed better accuracy and less error rate.