Mixed-effects and clustering approaches to thermal comfort in Indian university buildings

Assessing thermal comfort in naturally ventilated (NV) buildings is complicated by hierarchical data structures and limitations of temperature-centric models in warm, free-running contexts. This study reports a cross-sectional field survey in five NV university buildings in Hyderabad, India (hot semi-arid, BSh climate), across two monsoon seasons (N = 1,139). Operative temperatures (T op ) ranged from 26.1 to 33.8°C (mean 29.6°C). Linear mixed-effects models identified T op (β = 0.360), metabolic rate (β = 0.388), and season (β = 0.431) as significant predictors of thermal sensation; session-level variance exceeded building-level variance in a joint random-intercept model (ICC 0.163 vs. 0.020). Regression yielded a neutral temperature of 28.3°C (Griffiths mean 28.8°C at G = 0.50 K⁻¹; range 28.0–28.8°C across G = 0.25–0.50 K⁻¹), approximately 1.5–2.3°C warmer than ASHRAE Standard 55 predictions; the India Model for Adaptive Comfort (IMAC) and EN 16798-1 showed closer agreement. Principal component analysis revealed three perceptual axes (thermal load, air movement, humidity experience). K-means clustering (k = 3) tentatively identified three thermo-perceptual profiles, although moderate separation (silhouette = 0.235) suggests these should be treated as exploratory. Two groups at nearly identical operative temperatures (ΔT op = 0.17°C; Cliff’s δ = 0.14) differed substantially in thermal sensation (ΔTSV ≈ 1.03; δ = 0.43), associated with humidity and air-movement perception—revealing response-group divergence not captured by population-average regression. These patterns appear session-specific rather than stable occupant traits. Temperature-based comfort evaluation alone may overlook humidity- and air-movement-related perceptual heterogeneity in monsoon-season NV classrooms in BSh climates.