Applying the Gaussian Mixture Model to Generate Large Synthetic Data from a Small Data Set

Immersive virtual environments (IVEs) have been widely used as multipurpose tools in many research areas. As design tools, IVEs have shown the potential to observe human-building interactions for buildings under design. IVEs allow researchers or designers to understand human-building interactions and refine building functions to optimally fulfill design goals. However, due to limitations of the technology, practical use of IVE often result in small sample sizes, which may negatively impact those applications (e.g., machine learnings) that require a large amount of data. This paper demonstrates the application of a Gaussian mixture model (GMM) as a method for generating independent and identically distributed (IID) samples of data originally obtained from IVE applications to solve the issue of the small sample size associated with IVE experiments. In this study, GMM is tested using an application that involves light switch uses in a single office. First, an IVE of the office is created to simulate key artificial light use events during design. Then, light switch locations, tasks, work area illuminance, and factors potentially influencing light switch uses are modeled in the IVE experiment. Finally, thirty people participated in the experiment to collect data. The results of this study show that the IVE data and the IID samples are not significantly different, affirming that the GMM can be a good candidate for tackling a small sample size issue associated with IVE experiments.