Predicting the Student's Perceptions of Multi-Domain Environmental Factors in A Norwegian School Building: Machine Learning Approach

Abstract

Poor Indoor Environmental Quality (IEQ) in schools significantly impacts students’ well-being, learning capabilities, and health. Perceived dissatisfaction rates (PD%) among students often remain high, even when indoor environmental variables appear well-controlled. This study aimed to predict student dissatisfaction rates for multi-domain environmental factors using machine learning (ML) models by integrating building parameters, environmental data, and student dissatisfaction responses. The methodology involved collecting IEQ data from three classrooms using sensors, 1,437 student survey responses, and outdoor weather data, followed by statistical tests and ML modelling. Predicting PD% for all environmental factors using ML achieved reliable results (R² train > 0.81 & test > 0.74). SHAP analysis identified environmental variables and building parameters as critical contributors to perception of IAQ, thermal and visual environment. Random Forest (RF) as ML algorithm achieved the highest accuracy for overall IEQ dissatisfaction (PDIEQ%), with R² values of 0.91, outperforming Multi-Linear Regression (MLR), which had an R² of 0.56. IAQ, thermal comfort, and acoustic environment emerged as the most influential environmental domain factors for PDIEQ%. The findings emphasize the importance of considering both indoor and outdoor environmental factors, along with room-specific characteristics, to improve IEQ and reduce PD%. This study provides a framework for IEQ optimization using ML applications but is limited to a single school in a cold-climate region and a specific age group. Future research can expand to other climates, buildings, measurements, occupant levels, and other potential variables.

Keywords: Multi-domain Environmental Factors, User-Feedback, Machine learning, Random Forest