A Review of Machine Learning Methods in Building Energy Performance Assessment

Creative Commons License

Kunkcu H., KOÇ K., GÜRGÜN A. P.

7th World Congress on Civil, Structural, and Environmental Engineering, CSEE 2022, Virtual, Online, 10 - 12 April 2022 identifier

  • Publication Type: Conference Paper / Full Text
  • Doi Number: 10.11159/icsect22.174
  • City: Virtual, Online
  • Keywords: artificial intelligence, Building, energy consumption, energy performance, facility management, prediction, review
  • Yıldız Technical University Affiliated: Yes


© 2022, Avestia Publishing. All rights reserved.Buildings alone account for nearly one third of the world's total energy consumption. Therefore, it is of great importance to determine the energy performance of buildings with prediction models to contribute to the reduction in energy consumption. Machine learning (ML) methods have extensively been adopted in the assessment of the energy performance of buildings in the literature. This research aims to review existing studies in the literature to predict energy performance through ML methods. According to a comprehensive literature survey, 79 articles were identified and reviewed intensively based on several aspects such as journal, publication year, examined country, adopted programming tools, building types and utilized ML methods. Results of the literature survey indicated that different building types (i.e., residential or non-residential buildings) have different energy performances. Methodological investigation showed that artificial neural network and support vector machine methods were the most frequently implemented ML techniques in the prediction of energy performance. It was observed that many authors compared the performances of several ML methods to highlight the most capable methods. In addition, the energy performances of buildings were evaluated with ML methods using different programming tools in variety of countries. Overall, this study is expected to provide valuable information about the current state of ML methods to practitioners and researchers in this field.