Predicting energy consumption of building clusters at the design stage using machine learning models

The environmental impact of high energy consumption in buildings during the COVID-19 pandemic has led to the adopting of data-driven approaches for enhanced decision-making and energy savings. However, forecasting energy use during the early design phase remains limited. This study investigates how building clusters affect model performance at the design stage using five machine-learning techniques with a dataset of 10,264 buildings. Model performances were evaluated using their accuracy, RMSE, MAE, MSE, and R 2 metrics. Results showed that DNN achieved the best accuracy score of 98%, followed by MLPNN and SVM W with accuracy scores of 95% and 92%, respectively. The study proposes a general framework to predict average annual energy use across different building types at the early design stage, supporting informed and sustainable architectural decisions.