Classifying Floor Level Based on Human Comfort and Environmental Parameters Using Machine Learning Approaches for a Sustainable Indoor Analytics G.

As educational institutions evolve into complex structures, particularly in dense urban environments, understanding the vertical stratification of indoor environmental quality and related occupant comfort has become essential for sustainable building design. Thus, the primary objective for this research is to classify different floor levels based on human comfort and relating environmental parameters within the context of educational buildings’ lobby spaces located in Dhaka, Bangladesh, using machine learning (ML) models. In total, 265 data points were collected, including quantitative and qualitative at the winter season. For this analysis, two ML algorithms were tested: Random Forest (RF) classification and Decision Tree (DT) classification. Feature ranking and explainability are simplified using SHAP (Shapley Additive exPlanations) analysis. The RF model obtained a 79% accuracy score, with temperature, humidity, density of people, and age being the most important features. The DT model had the most accuracy at 81%, with temperature, humidity, density of people, and lighting level being the most important features. These findings underscore the ML models’ performance in lobby spaces across different floor levels, highlighting key features. The study contributes to the advancement of interpretable ML approaches for floor level classification, offering actionable insights for the design of occupant-centric and energy-efficient educational buildings. Exploring the Role of Peri-urban Areas in Driving Self-help Housing: A Case Study of Tea Estate,