Prediction of Homeowners' Renovation Tendency Based on Multi-Head Attention Mechanism

Research on homeowners’ renovation tendencies remains scarce, particularly in the development of quantitative and predictive methods, while current market promotion in the renovation industry practices rely heavily on broad advertising or passive customer inquiries, leading to low targeting efficiency and high operational costs. To address this gap, this study introduces a methodological framework to quantitatively assess and predict the likelihood that households will undertake renovation projects. We first construct a set of well-defined indicators to measure renovation tendency, capturing demographic, behavioral, and contextual factors, thereby addressing the lack of standardized predictive metrics in the field. These indicators are then incorporated into a predictive model that combines the Multilayer Perceptron with a multi-head attention mechanism, enabling the model to capture complex and non-linear relationships within the data. Experimental results demonstrate that the constructed indicator system can effectively predict homeowners’ renovation behavior, achieving an accuracy of 58.90 %–82.41 %. Beyond offering a critical academic insight for analogous prediction tasks in housing and construction research, this research offers a solid theoretical foundation and scalable, practical tools for development of the building renovation industry and the formulation of urban regeneration policies.