Simulating Events to Generate Synthetic Data for Pervasive Spaces

Actual data collected from real deployments is ultimately the best data that can be had and used in evaluating systems and new concepts. However, available data may not be specific enough to drive certain evaluation goals. Therefore, it is necessary to propose, as an alternative, a method to generate synthetic data for pervasive spaces. A first step in this direction is trying to simulate a chain of events in the sensor output space by the use of pattern recognition models. Specifically, we propose the use of Markov Chains to generate patterns of events due to the fact that they have been used successfully to classify chains of events in machine learning and computer vision. We believe this strategy will make for a more realistic simulation of sensor activity in a pervasive space.