This paper is concerned with the development of data-driven predictive models capable of forecasting commercial building heating loads based on BEM (Building Energy Management) systems recorded variables, as well as weather data. To address the lack of available complete datasets from actual commercial building BEM systems, a detailed representation of a reference building using EnergyPlus was implemented as a benchmark. Data analysis of the simulated results is used to detect relationships between variables and select input variables for the predictive models. Various regression and machine learning models are investigated for their ability to forecast building heating loads. The most suitable model is selected by comparing the accuracy of the predictions.