Best subset selection multiple linear regression model for energy consumption prediction in educational buildings.

Baseline Energy Models developed using Multiple Linear Regression (MLR) have been widely adopted due to their simplicity in representing energy consumption and its related independent variables. Moreover, MLR provides easily interpretable insights into the possible independent variables that govern energy consumption. Nonetheless, the risk of overfitting or underfitting may occur when unnecessary or correlated independent variables are included in the model. The objective of this paper is to enhance the selection of independent variables through the Best Subset Selection (BSS) method applied to baseline energy models developed using MLR. Two educational buildings were used as case studies. Baseline energy models were developed using both MLR and MLR enhanced with BSS, and the results of model development and prediction accuracy were compared. In case building 1, the MLR model enhanced with BSS showed improved prediction accuracy of Mean Square Error and, Root Mean Square Error of 255.0059kWh and 15.9689 kWh respectively. However, in case building 2, the enhanced model did not improve prediction accuracy. Nonetheless, in case building 2, the differences in accuracy between the enhanced and non-enhanced models were minimal, indicating that building owners may adopt either the standard or enhanced model with confidence when developing baseline energy models.