Unsupervised context discovery based on hierarchical fusion of heterogeneous features in real smart living environments.

The development of Internet of Things (IoT) has enhanced smart living environments with great support by heterogeneous sensors for many human-centric purposes, among which an important one is to recognize potential contexts through Big Data analytics. In the related researches, the analytics is usually conducted on activity data from well-designed smart environments, where sensors are distributed in a balance manner for specific activity patterns predefined by domain knowledge. However, in a real living environment, sensor deployment is usually ad hoc customized without considering specific activity patterns in advance. For large quantity of data, it would be desirable to conduct unsupervised data-driven analytics for context discovery. Moreover, features from heterogeneous sensors may conflict with one another, and conventional methods often ignore the background context, which usually exists in reality and compromises the discovery of minor contexts. Therefore, in this paper, we propose an unsupervised analytics framework to discover the potential daily contexts for real smart living environments based on hierarchical fusion of features from heterogeneous sensors. The experimental results show that the contexts discovered by our proposed work present a more reasonable way to describe our daily lives.