Bolluk, Muhammed SaidSeyis, SenemAydoğan, Reyhan2024-02-162024-02-162023978-070170273-12684-1150http://hdl.handle.net/10679/9158https://doi.org/10.35490/EC3.2023.291Building energy demand assessment plays a crucial role in designing energy-efficient building stocks. However, most studies adopting a data-driven approach feel the deficiency of datasets with building-specific information in building energy consumption estimation. Hence, the research objective of this study is to extract new features within the climate, demographic, and building use type categories and increase the accuracy of a non-parametric regression model that estimates the energy consumption of a building stock in Seattle. The results show that adding new features to the original dataset from the building use type category increased the regression results with a 6.8% less error and a 30.8% higher R2 Score. Therefore, this study shows that building energy consumption estimation can be enhanced via new feature extraction equipped with domain knowledge.engrestrictedAccessconferenceObject10.35490/EC3.2023.291Machine learningUrban building energy demand predictionFeature extraction2-s2.0-85177215718