Application of ANN-ML in evaluating indoor thermal comfort of trio-typology residential building

Abstract

Rapid urbanization in regions like Abuja, Nigeria, poses significant challenges in achieving sustainable indoor thermal comfort, particularly when integrating traditional building techniques. This study evaluates adaptive thermal comfort in trio-typology residential buildings (TTRB), namely; modern (BDG1), traditional (BDG2), and compressed earth (BDG3), using an artificial neural network machine learning (ANN-ML) model. Environmental parameters were monitored over seven days during the rainy and dry seasons, alongside 230 occupant surveys capturing thermal sensation (TSV) and preference (TPV) votes. A simulation of a three-bedroom model was conducted using Design Builder software to assess natural ventilation (NV) performance. The ANN-ML model predicted TSV and thermal comfort (Tc) at 70 % training, 15 % validation, and 15 % testing under controlled conditions.

Results revealed rainy season temperatures ranged from 23.6 °C to 29.3 °C, while dry season temperatures spanned 25.0 °C to 35.3 °C. Thermal comfort zones were established at 27.4 °C (rainy season) and 20.54 °C (dry season). During the rainy season, 47.3 % of occupants reported feeling “slightly cool,” while 33.6 % felt “slightly warm.” in the dry season, 41.8 % perceived morning conditions as “cool,” with 63.6 % preferring warmer temperatures. The ANN-ML analysis demonstrated high accuracy in predicting occupant thermal comfort in NV buildings, confirming its reliability for such assessments. The findings highlight the potential of traditional building techniques to significantly enhance indoor thermal comfort. This study underscores the importance of integrating passive design strategies and machine learning models to optimize sustainable building performance in tropical climates.