Energy-balanced sensor selection for social context detection

Context detection is essential in pervasive computing to adapt the application behaviour to the user's context, like location, activities, social relationship. The key challenge is to efficiently determine context with a high accuracy based on low level sensor data. In this paper, we define the problem to maximize quality of information (QoI) of context detection with budget constraint for all available sensors and for groups of sensors, i.e., those that are on the same mobile phone. We formulate this problem as multi round sensor selection problem and show it to be NP complete. We propose a brute force and a heuristic method and show through simulation the effectiveness of our methods to extend the system life time and guarantee QoI for longer time. Furthermore, we show that the QoI of our heuristics is very close to the QoI of the brute force method, but its computational complexity is orders of magnitude smaller and as such suitable for real time applications.