© 2022, Springer Nature Singapore Pte Ltd.Population growth in cities negatively affects global climate problems regarding environmental impact and energy demand of building stock. Thus, buildings should be examined for energy efficiency by reaching acceptable internal thermal comfort levels to take precautions against climate disasters. Although building energy simulations (BES) are widely used to examine retrofitting processes, the computational cost of urban-scale simulations is high. The use of machine learning techniques can decrease the cost of the process for the applicability of quantitative simulation-based analyses with high accuracy. This study presents the implementation of the k-means clustering algorithm in an Urban Building Energy Modeling (UBEM) framework to reduce the total computational cost of the simulation process. Within the scope of the work, two comparative analyses are performed to test the feasibility of the k-means clustering algorithm for UBEM. First, the performance of the k-means clustering algorithm was tested by using the observations on the training data set with design parameters and performance objectives. The second analysis tests the prediction accuracy under different selection rates (5% and 10%) from the clusters partitioned by the k-means clustering algorithm. The predicted and simulation-based calculated results of the selected observations were comparatively analyzed. Analyses show that the k-means clustering algorithm can effectively build performance prediction with archetype characterization for UBEM.