Data-driven Machine Learning and 3D Visualization for Forecasting Indoor Environmental Conditions A Case Study on Campus Building (Aule R, Polito)

Indoor environmental conditions, particularly air temperature and CO₂ concentration, play a central role in shaping comfort, health, and energy performance in academic buildings. Despite the widespread deployment of IoT-based monitoring systems, most HVAC systems still operate reactively. This thesis addresses this gap by developing an explainable, data-driven forecasting framework that predicts short-term indoor environmental trends and integrates these predictions into a 3D BIM-based visualization environment for improved decision support. Using one year of sensor data from the Aule R building at Politecnico di Torino, the study evaluates the performance of both machine-learning and deep-learning methods, including Random Forest, CNN, and LSTM for modeling temperature and CO₂ dynamics. The methodology includes a pipeline of data cleaning, temporal alignment, feature engineering, multivariate time-series forecasting, uncertainty estimation. The framework connects predictive outputs to an innovative dashboard, enabling spatial representation of forecasted indoor conditions within the building’s 3D model. The expected outcome is that machine-learning models can provide reliable short-horizon temperature forecasts by leveraging the temporal structure inherent in indoor sensor data. Furthermore, explainability tools help clarify the role of key variables, while the BIM-enabled data visualization aims to offer an intuitive platform for exploring environmental trends and supporting proactive HVAC-related decision-making.