Self-adaptive and lightweight real-time sleep recognition with smartphone

Abstract

It is widely recognized that sleep is a basic physiological process having fundamental effects on human health, performance and well-being. Such evidence stimulates the research of solutions to foster self-awareness of personal sleeping habits, and correct living environment management policies to encourage sleep. In this context, the use of mobile technologies powered with automatic sleep recognition capabilities can be helpful, and ubiquitous computing devices like smartphones can be leveraged as proxies to unobtrusively analyse the human behaviour. To this aim, we propose the implementation of a real-time sleep recognition methodology relied on a smartphone equipped with a mobile app that exploits contextual and usage information to infer sleep habits. As an improvement of already presented solutions, in this proposed application an initial training stage is required, during which the selected features are processed by k-Nearest Neighbors, Decision Tree, Random Forest, and Support Vector Machine classifiers, in order to select the best model for each user. Moreover, a 1st-order Markov Chain is applied to improve the recognition performance. Experimental results demonstrate the effectiveness of the proposed approach, achieving acceptable results in term of Precision, Recall, and F1-score.