Abstract
<title>Abstract</title> <p>Residential energy efficiency is now an integral part of climate change mitigation and sustainable urban planning. Despite recent developments in the artificial intelligence sector, conducted on multi-criteria based assessments of housing energy efficiency possible, the studies focus more on building or environmental features than on the socio-demographic characteristics of occupants and their impacts on the energy performance of housing units. This gap reduces the possibilities of an occupant-responsive and personalized energy advisory platform. This is especially important for apartment housing, which has a great share of energy consumption, carbon emission, and heat generation in many large cities, occupants are mostly ordinary people who are absent during the housing provision process. To tackle this gap, this study presents and tests an AI-based human-in-the-loop framework for multi-criteria evaluation of housing energy efficiency and for providing tailored energy-efficiency advice on the basis of households’ characteristics in apartment housing in Mashhad, Iran. This study brings together key socio-demographic data, building physical characteristics, neighbourhood environmental features, and climatic information in a single analytical platform for residential energy assessment to provide energy-efficiency recommendations to the individual. Data from 50 apartment households in 20 apartment buildings located in Mashhad were analysed using a multi-stage analytical process, including reliability assessment, statistical data description, random forest regression, analysis of feature importance, and five-fold cross validation. The proposed framework yielded a coefficient of determination (R²) of 0.86, mean absolute error (MAE) of 0.14 kWh/m² and the root mean squared error (RMSE) of 0.18 kWh/m², which indicate satisfactory predictive performance in the context of a proof of concept implementation. Feature importance analysis was used to identify the main features that impact residential energy performance, such as natural ventilation, building age, neighbourhood green cover, and occupant age. Moreover, the coupling of predictive modelling with rule-based reasoning showed a promising way to produce interpretable, personalized energy efficiency recommendations based on occupants’ characteristics. The results indicate that the energy efficiency of the residential building is determined by the synergistic effect of the occupant, the building, the neighbourhood and the climatic data, and not just by the building. The proposed framework serves as a solid basis for the future development of human-centred, AI-supported housing energy advisory platforms that support users, designers, planners, and policy makers towards more energy-efficient, user-friendly and sustainable housing environments.</p>