UCLA

Nowadays, most commercial and residential buildings are instrumented with a variety of meters and sensors as part of a utility infrastructure installed by service providers. This thesis explores the extent to which readily available sensor information may be used to make occupancy-related inferences. Particularly, we focus on inferences which can be made from smart electric meters, water flow sensors, and network traffic monitors and have gathered several months worth of relevant data from two distinct settings. We explore various machine learning techniques to evaluate a service provider's ability to make privacy-invasive inferences. Our results show that, even with coarse-grained sensor data, we are able to make occupancy related inferences significantly better than naive prediction strategies where sensor information is not taken into account.