A comprehensive review and future research directions of ensemble learning models for predicting building energy consumption

Abstract

Ensemble learning has garnered increasing attention in building energy prediction over the past decade due to its exceptional predictive accuracy. However, there is a lack of systematic reviews that comprehensively analyze its current research status, limitations, and challenges, particularly in the context of large-scale practical applications. To address this gap, this review article systematically evaluates the application of ensemble learning models in building energy prediction. Using the PRISMA method, 82 relevant articles published between 2013 and 2024 in the Web of Science database were analyzed. The findings indicate that heterogeneous ensemble models, which integrate diverse algorithms, and homogeneous ensemble models, which utilize multiple data subsets, both hold significant potential for enhancing prediction accuracy. Specifically, heterogeneous models achieved accuracy improvements ranging from 2.59% to 80.10%, while homogeneous models demonstrated more stable improvements of 3.83% to 33.89%. Nonetheless, the integration of multiple base models increases computational complexity, resulting in higher computation times. Despite this drawback, the improved prediction accuracy, robustness, and generalization capabilities of ensemble models justify the additional computational cost. The review identifies key limitations, including the subjective selection of learning algorithms, the lack of systematic methods for evaluating model diversity, and insufficient exploration of combination strategies. Future research should focus on developing objective criteria for algorithm selection, advancing diversity evaluation techniques, analyzing the effects of combination methods, comparing computational efficiency, and validating the robustness and generalizability of ensemble models. This study offers valuable insights for researchers and practitioners aiming to optimize ensemble learning models in building energy prediction.