PropTech Blogs and Articles

When Data Handling Alters Physical Interpretation: HVAC’s Missing Evidence Layer

A 2026 Energy and AI study shows that HVAC thermal‑comfort models produce wildly different variable importance ratios when missing data is handled differently, shifting the air‑temperature to mean‑radiant‑temperature (Ta:MRT) ratio from 1.9:1 to 4.46:1 – a 135% change. The researchers used the same dataset and LightGBM model, demonstrating that data imputation can create artificial correlations and suppress radiant effects. The most accurate models, judged by prediction error, were also the most misleading in physical interpretation. The authors argue that the industry must move from a data‑centric to a truth‑centric workflow that preserves complete environmental evidence at capture.