IoT-Based Indoor Thermal Comfort Prediction Using Multivariate Statistical Analysis

Emerging Internet of Things (IoT) technologies, combined with sensors and data analytics, enable intelligent management of buildings, significantly improving energy efficiency and optimizing thermal comfort for occupants. In this article, we present a novel modeling framework that integrates IoT-based monitoring with multivariate statistical analysis (MVA) to assess indoor thermal comfort and predict the optimal temperature (T) that maximizes occupant satisfaction within a given space. By continuously monitoring real-time environmental conditions, 11 indoor and outdoor variables are collected every 20 min from three different locations via an IoT network. Additionally, surveys on occupant satisfaction and thermal comfort are conducted four times daily over 21 days to evaluate overall indoor comfort. Both linear models, using standard regression and nonlinear models, adopt locally weighted regression (LWR) methods, which were exploited for the first time in this context to predict indoor comfort using current and historical environmental data. The model achieves a prediction error of 9.6% in predicting user satisfaction at the selected site. Another predictive model is developed to predict the optimal T that maximizes occupant satisfaction. Notably, the model demonstrates significant improvements in occupant satisfaction across three different rooms, increasing from 5.18% to 100%, 12.69% to 91.53%, and 58.48% to 100%, respectively. This study highlights the significant role of IoT and MVA in uncovering intricate correlations within complex datasets. To our knowledge, this is the first time MVA has been applied to indoor wellness models, where neural networks (NNs) are currently the most used approach with higher computational complexity. The proposed framework sets the stage for the optimum control of heating, ventilation, and air conditioning (HVAC) systems in buildings. By leveraging IoT and MVA, this framework provides actionable insights for HVAC system optimization, reducing unnecessary energy consumption while maintaining occupant comfort. This research demonstrates a practical, interpretable, and scalable alternative to NNs-based models, making it highly suitable for real-time building management and smart climate control systems.