Abstract

The integration of waste electrical and electronic equipment (WEEE) into construction materials represents a promising circular economy strategy for sustainable building design. However, the indoor hygrothermal performance of such eco-friendly buildings under real climatic conditions remains insufficiently understood. This study investigates the indoor temperature and humidity behavior of buildings incorporating different proportions of WEEE-derived glass materials and develops a predictive framework for indoor environmental conditions. Three experimental buildings were constructed: a control building (C100PV0) and two eco-friendly configurations, C70PV30 (30% glass powder) and GV50Sa50 (50% glass aggregate). Indoor and outdoor data were collected using IoT-based sensors combined with external NASA atmospheric data. An XGBoost model incorporating lagged and rolling features was developed and evaluated using a chronological train–test split to ensure realistic validation. The results show that indoor environments exhibit reduced variability, reflecting the thermal and hygrometric damping effect of construction materials. The C70PV30 building demonstrates the highest thermal stability, while GV50Sa50 exhibits the most stable humidity evolution with minimal fluctuations. From a predictive perspective, the model achieves excellent performance for temperature in both buildings and for humidity in C70PV30 (R² ≈ 0.99), with error levels of RMSE ≤ 0.10 for temperature and ≤ 0.22 for humidity. In contrast, humidity prediction in GV50Sa50 yields a low R² (≈ 0.12) due probably to the near-constant nature of the signal; however, error-based metrics (MedAE ≈ 0.04) confirm that predictions remain accurate in absolute terms. Overall, the findings demonstrate that incorporating recycled glass materials improves indoor hygrothermal stability while maintaining high predictability. The proposed IoT–machine learning framework provides a robust approach for modeling and optimizing indoor environmental conditions in sustainable buildings.

This work is licensed under a Creative Commons Attribution 4.0 License.