Style classification and prediction of residential buildings based on machine learning

Abstract

ABSTRACT Architectural style positioning is an important part in the process of residential building design and project planning. However, in practice, due to the complexity and ambiguity of styles, style positioning often relies more on the subjective judgement of the designers and lacks scientificity. This paper proposes a method for the classification and prediction of residential building styles. Through structured interviews and questionnaire surveys on front-line designers and project planners, it refines the key morphological elements and the site economic factors that influence architectural style classification and positioning. Based on machine learning, after analysing the data of 372 newly built real estate projects in Hangzhou, the research finds t|TABE_A_1779728|TABE_A_1779728hat the current real estate styles can generally be divided into 8 categories. Whether it is a curved volume, the shape of the roof and the richness of the tones are the most important morphological variables that differentiate style categories, and the building height is the most important economic factor for style positioning. When using the selected five economic factors as independent variables to train a neural network model and predict the morphological elements and style categories, the average accuracy reaches 77.2%.