engleski

Human activity recognition with smartphones and its economic implications

This paper suggests models for human activity recognition with the use of artificial neural networks and decision trees. Today, smartphones are very popular which makes the use of smartphone sensors most suitable and cost-effective tools for measuring human activities. Human recognition technology can be used in healthcare (recording data about elderly people and preventing their falls etc.), in recording people’s everyday activities, location etc. The purpose of this paper is to model human activity recognition through several goals: (1) to describe technology, types and approaches of human activity recognition, the challenges of its applications, and (2) to create a machine-learning model that will be able to recognize human activity by using artificial neural networks and decision trees. The dataset on human activity recognition from the website Kaggle.com was used, which consisted of more than five hundred input variables and the output variable expressed in six categories: laying, sitting, standing, walking, walking downstairs and walking upstairs. Both neural network and decision tree methods were trained on the same training subset and their accuracy on the same test subset was measured and compared. Besides testing model accuracy, the importance of input variables was also analyzed. The results showed that the most accurate model was obtained by neural networks with 74.45% of accurately recognized human activities, whereas the decision tree model accuracy was 67.88%. Possible applications of human recognition technology are numerous, like personal biometric signature, care of elderly people and infants, localization, industrial production etc. There are certain limitations of using this type of technology like user sensitivity (due to different ways of doing the same movements), location sensitivity, the complexity of the activity (if users do many activities at the same time), energy limitations, source limitations, an insufficient number of subjects for training.

decision trees, human activity recognition with smartphones, human activity recognition technologies, neural networks, sensors in smartphones

nije evidentirano