Approximator: Predicting Interruptibility in Software Development with Commodity Computers

Assessing the presence and availability of a remote colleague is key in coordination in global software development but is not easily done using existing computer-mediated channels. Previous research has shown that automated estimation of interruptibility is feasible and can achieve a precision closer to, or even better than, human judgment. However, existing approaches to assess interruptibility have been designed to rely on external sensors. In this paper, we present Approximator, a system that estimates the interruptibility of a user based exclusively on the sensing ability of commodity laptops. Experimental results show that the information aggregated from several activity monitors (i.e., Key-logger, mouse-logger, and face-detection) provide useful data, which, once combined with machine learning techniques, can automatically estimate the interruptibility of users with a 78% accuracy. These early but promising results represent a starting point for designing tools with support for interruptibility capable of improving distributed awareness and cooperation to be used in global software development.