Study of Multi-Classification of Advanced Daily Life Activities on SHIMMER Sensor Dataset

Abstract

Today the field of wireless sensors have the dominance in almost every person’s daily life. Therefore researchers are exasperating to make these sensors more dynamic, accurate and high performance computational devices as well as small in size, and also in the application area of these small sensors. The wearable sensors are the one type which are used to acquire a person’s behavioral characteristics. The applications of wearable sensors are healthcare, entertainment, fitness, security and military etc. Human activity recognition (HAR) is the one example, where data received from wearable sensors are further processed to identify the activities executed by the individuals. The HAR system can be used in fall detection, fall prevention and also in posture recognition. The recognition of activities is further divided into two categories, the un-supervised learning and the supervised learning. In this paper we first discussed some existing wearable sensors based HAR systems, then briefly described some classifiers (supervised learning) and then the methodology of how we applied the multiple classification techniques using a benchmark data set of the shimmer sensors placed on human body, to recognize the human activity. Our results shows that the methods are exceptionally accurate and efficient in comparison with other classification methods. We also compare the results and analyzed the accuracy of different classifiers.