Occupancy inferencing from non-intrusive data sources

Intuitively, measurements from utility meters that are associated with a physical space have embedded in them some information about the occupants of that space. Occupancy information can be sensitive yet empowering. On one hand, with the right information, administrators can adjust subsystems to maximize comfort and energy efficiency. On the other hand, sensitive details about occupants may be leaked. We explore the accuracy to which meter data from physical spaces, when subjected to machine learning algorithms, can yield occupancy information. Our results can then be used to devise low-cost mechanisms for occupancy sensing from the opportunistic use of already available data, and to quantify the risk of leaking privacy-sensitive inferences.