Feature Importance Analysis of Thermal and Visual Comfort in Urban Slum Areas Using Artificial Neural Networks Md. Faisal

Urban regions with inadequate infrastructure and poor living conditions, known as slums, are rapidly expanding in low-income cities like Dhaka, Bangladesh, due to urbanization, economic constraints, and high housing costs. Although previous studies have generally examined these comfort dimensions separately, limited attention has been given to identifying the key factors influencing both simultaneously in slum environments. This study develops machine leaning (ML) model: Artificial Neural Network (ANN); to identify and rank the factors affecting thermal and visual comfort (TVC) using four feature selection techniques: Random Forest (RF), Principal Component Analysis (PCA), Lasso Regularization (LR), and Recursive Feature Elimination (RFE). A dataset comprising 400 winter-season observations and 25 demographic, environmental, behavioral, and housing- related variables was collected through field surveys and environmental monitoring. The optimized ANN models achieved an overall classification accuracy of 0.76 for both TVC prediction. While the four feature selection methods produced different rankings because they evaluate feature importance from different perspectives, several variables, including illuminance (lux), TVOC, humidity, kitchen location, clothing level, occupant density, and current location, consistently appeared as influential determinants. These findings provide practical evidence for occupant-centered strategies and support sustainable, climate-responsive housing design and environmental management in informal settlements. Human Dynamics Influencing Indoor