The aesthetic attractiveness of urban façades exerts a considerable influence on the overall quality of urban environments and the satisfaction of their inhabitants. Nevertheless, the diverse and non-quantifiable interpretations of beauty continue to pose an enduring challenge for architectural designers. This study aims to establish a predictive intelligent model for the quantitative assessment of aesthetic quality in urban building façades. Accordingly, a dataset was provided, including 77,600 professional evaluations obtained through a structured survey administered to 388 active architects and urban designers. To strengthen generalizability, 1,300 façade images from various cities representing different climates and cultural contexts were selected and subsequently evaluated by specialists. The most influential factors of façade aesthetic qualities, as identified by specialists, were derived through an open-ended questionnaire, qualitative analysis techniques, and the utilization of MAXQDA software. Next, the well-known deep learning algorithms, such as EfficientNet, VGG16, ResNet, YOLOv8, and MobileNet, were trained on 70% of the dataset, evaluated on 20%, and tested on the remaining 10%. The results showed that the YOLOv8 model exhibited superior performance, with an accuracy of 81.9% in forecasting expert assessments of the aesthetic quality of building façades. This study advances the quantitative evaluation of aesthetic features of urban building façades and provides a suitable method for analyzing the aesthetics of façades before their implementation.
