Benchmarking Classical Machine Learning and Hybrid Quantum Approaches for Indoor Thermal Sensation Prediction
Indoor Air, cilt.2026, sa.1, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 2026 Sayı: 1
- Basım Tarihi: 2026
- Doi Numarası: 10.1155/ina/3385686
- Dergi Adı: Indoor Air
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Compendex, Environment Index, Academic Search Ultimate (EBSCO), Natural Science Collection (ProQuest), Biomedical Reference Collection: Corporate Edition (EBSCO), Health Research Premium Collection (ProQuest)
- Anahtar Kelimeler: ASHRAE, machine learning, occupant-centric modeling, quantum machine learning, thermal comfort prediction
- Süleyman Demirel Üniversitesi Adresli: Evet
Özet
Reliable prediction of thermal sensation plays a critical role in enabling energy-efficient and occupant-oriented building operation. However, existing approaches remain limited under real-world conditions. This study establishes a comprehensive and methodologically consistent framework to evaluate the predictive performance of thermal sensation vote (TSV) models by integrating a broad set of classical machine learning algorithms with a hybrid quantum–classical approach. The analysis is based on a dataset of 19,291 observations from naturally ventilated environments, including both environmental and personal variables. A total of 14 regression models were benchmarked using a unified fivefold cross-validation protocol, followed by hyperparameter optimization of the best-performing model and a structured feature ablation analysis. In parallel, a hybrid quantum–classical model was developed using parameterized quantum circuits and evaluated under the same experimental conditions to ensure direct comparability. The results show that ensemble-based classical models outperform all other approaches, with tuned XGBoost achieving the highest predictive performance (R2: 0.386 and RMSE: 0.983). However, improvements from hyperparameter optimization remain marginal, indicating a practical predictive ceiling. The feature ablation analysis reveals that personal variables provide the most significant contribution to prediction accuracy, while derived thermal indicators such as operative temperature offer no additional benefit when their constituent variables are included. The hybrid quantum model demonstrates stable training behavior but significantly lower predictive performance (R2: 0.246 and RMSE: 1.085) and early convergence, suggesting limited representational capacity for tabular and noisy thermal comfort data. The findings indicate that TSV prediction is fundamentally constrained by data characteristics and human variability rather than model complexity. The study highlights the critical importance of occupant-related variables, identifies diminishing returns from conventional feature engineering, and demonstrates that current quantum machine learning approaches are not yet competitive for this problem domain. These results provide a robust benchmark and emphasize the need for data-centric and human-centric modeling strategies to advance thermal comfort prediction.