Rahmanparast A., Milani M., Dalkılıç A. S.
Scientific works/Elmi eserler, ss.153-160, 2026 (Hakemli Dergi)
Özet
nsuring user comfort in buildings is a
crucial research topic in terms of both indoor
environmental quality and energy efficiency. This
study aims to classify users' thermal comfort status
using machine learning methods. Using Class I
data from the Chinese Thermal Comfort Dataset,
the thermal comfort vote problem was transformed
into a binary classification problem.
Environmental, personal, and building-related
variables were used as model inputs; Logistic
Regression, Random Forest, and Gradient
Boosting algorithms were compared using
accuracy, precision, recall, F1-score, and ROCAUC metrics. According to the results, although
Random Forest provided the highest accuracy
(0.709) and ROC-AUC (0.768) values, the learning
curves showed an overfitting tendency in this
model. The Gradient Boosting model produced the
highest F1-score (0.789) and showed more
consistent generalization performance. Overall, the
results show that machine learning methods can be
used as an effective tool in predicting user thermal
comfor