Studies indicate that a large energy saving can be realized by applying automatic fault detection and diagnosis (AFDD) to building systems, which consumes more than 40% of the primary energy in the U.S. To enable AFDD, a baseline depicting the normal operation mode is needed to detect whether the building operation deviates from normality. Different from many other systems, a building system behaves differently under different weather conditions and hence needs its baseline model to reflect such weather dependence. Existing research shows baseline constructed using nonlinear mathematical models has performed well. However, determining the sample size, which is necessary to capture the totality of data space needed for accurate baseline construction, relies on trial-and-error experiments conducted offline. There is a lack of easy-to-use method that can provide guidance on whether enough sample size has been collected for baseline construction. In this research, we have developed a data-driven approach for AFDD baseline model construction based on information entropy, in conjunction with enthalpy, a measurement of outdoor air conditions to reflect weather conditions. The developed method is compared with our previously-reported baseline construction method using real building data.