Machine Learning Modeling for Building Energy Performance Prediction Based on Simulation Data: A Systematic Review of the Processes, Performances, and Correlation of Process-Related Variables
Machine learning models have become a potential alternative for building energy performance studies since they provide fast and reliable prediction results. However, decisions in the modeling process are sometimes made without knowing their possible impact on the results, which may lead to unstable...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2025-04-01
|
| Series: | Buildings |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2075-5309/15/8/1301 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Machine learning models have become a potential alternative for building energy performance studies since they provide fast and reliable prediction results. However, decisions in the modeling process are sometimes made without knowing their possible impact on the results, which may lead to unstable process management. Therefore, this study aims to obtain a machine learning modeling process framework focusing on critical-decision subjects through a systematic review of the recent literature. Preferences of the current supervised modeling practices on process-related variables to have prediction models with high accuracy were analyzed in the studies using simulation data. In this paper, a general framework of the processes is presented through their steps and decision subjects. Regarding these steps, the frequency of the methods used, strategies followed against the limitations, common sources of concerns, and intertwined workflows are analyzed with their effects on prediction performance in terms of accuracy. In addition, correlations between process-related variables, i.e., decision subjects and model performance, are investigated to quantify the impacts. As a result, the most effective decision subjects on accuracy were observed as the machine learning algorithm to be used, input variables to be included, and the range of the sample size, respectively. |
|---|---|
| ISSN: | 2075-5309 |