Accurate predictions of energy consumption are essential to optimizing building energy use performance. To date, substantial efforts have been undertaken to improve prediction accuracy, specifically while focusing on occupants' presence in buildings. Unfortunately, two significant obstacles remain when predicting building energy consumption using occupancy data. First, occupancy diversity among end-user groups is rarely considered during model development. Second, occupancy's correlation with energy consumption may be weak due to variances in occupant behavior. Therefore, this research aims to investigate how occupancy-related characteristics of end-user groups affect prediction performance. In order to achieve this objective, a data mining-based prediction model is constructed to mimic building thermal behaviors. The experimental results using the proposed prediction model make it evident that prediction accuracy is improved when considering diverse occupancy and its correlation with energy use. In addition, significant prediction accuracy is achieved using only a minimal amount of historical data. With the proposed prediction model, it is possible to obtain more detailed information about energy use patterns (e.g., load shape, the amount of energy use) for end-user groups. Thus, facility managers will be able to personalize the operation of energy-consuming equipment depending on end-user group for reducing energy consumption without compromising occupants' thermal comfort. (C) 2017 Published by Elsevier B.V.