In the effort to build a sustainable society, smart home research attention is being directed toward green technology and environmentally-friendly building designs. In this paper, we analyze the distribution of home energy consumption, and then present both linear and non-linear regression learning models for predicting energy usage given known human behavior and time-scale features. To guarantee the validity of our methods, two real-world data sets collected over three months are applied into training the models. Based upon our learning models, a web-based end-user system is developed for providing users feedback about behavior-based energy usage to promote energy efficiency and sustainability through behavior changes.