Feature Ranking for Predicting Occupant Thermal Comfort Using Artificial Neural Networks (ANNs)

Indoor comfort refers to the overall satisfaction and well-being of occupants in terms of thermal and visual conditions within the building. This study utilizes Artificial Neural Networks (ANNs) to predict occupant thermal comfort in a naturally ventilated educational building situated in Dhaka, Bangladesh. The primary aim is to identify the most significant features or feature rankings that have a substantial impact on occupant thermal comfort. Four feature selection methods, namely Principal components analysis (PCA), Tree-based (Random Forest), Recursive Feature Elimination (RFE), and Lasso regularization, were employed to assess feature importance and rankings. The results of the feature ranking analysis consistently highlight certain features as influential across the different approaches. Notably, as Floor Area, No of Windows, Lighting Level, Study Level, and CO₂ emerged as significant factors in predicting occupant thermal comfort. Additionally, features such as “Temperature”, “Humidity”, “Room Orientation”, “No of Fans”, and “No of Lights”, demonstrated varying degrees of significance. These findings provide valuable insights into the factors that contribute to occupant thermal comfort in the context of a naturally ventilated educational building. By understanding the optimal features or feature rankings, stakeholders can make informed decisions and implement strategies to enhance indoor comfort conditions.