Loading…

Geo-specific development of thermal comfort prediction models: A machine learning approach using the ASHRAE dataset

Thennal comfort is a crucial aspect that measures people’s satisfaction with their thennal sensations. It has a significant impact on individuals, affecting their work performance, cognitive function, and healthcare. Furthermore, interior designers may have difficulty determining the comfortability...

Full description

Saved in:
Bibliographic Details
Main Authors: Tan, Hui-Hui, Tan, Yi-Fei, Tan, Wooi-Haw, Ooi, Chee-Pun
Format: Conference Proceeding
Language:English
Subjects:
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Thennal comfort is a crucial aspect that measures people’s satisfaction with their thennal sensations. It has a significant impact on individuals, affecting their work performance, cognitive function, and healthcare. Furthermore, interior designers may have difficulty determining the comfortability of a space without expertise in thermal comfort assessment. Over the years, researchers have explored various approaches to develop thermal comfort models, including using the heat balance equation in the past. In recent years, with the rapid growth of machine learning technology, researchers have become more interested in using machine learning to develop thermal comfort models. Hence, in this paper, some machine learning models are selected to develop the thermal comfort model using a well-known thermal comfort dataset, namely ASHRAE Comfort Database II (version 2018) to predict 3-point Thermal Sensation Vote. The study introduces a novel approach to build thermal comfort prediction models that include country as a contributing factor. This new method has shown improvement compared to past research. The Random Forest method demonstrated significant potential for handling and performing well with datasets from various countries. The study unveiled only slight differences in predictive accuracy across various countries, regardless of using 6, 7, or 12 feature inputs, suggesting that increasing the number of feature inputs beyond seven did not significantly enhance the model’s performance. Furthennore, the models used in the study were deemed fair and reliable, supported by K-Fold cross-validation and minimal variation in the model results.
ISSN:0094-243X
1551-7616
DOI:10.1063/5.0240460