Predicting Potential Digital Technology Innovation in the Construction Industry: A Two-Stage Predictive Model Based on Machine Learning

Abstract Over the past decade, the construction sector has increasingly adopted digital technologies to improve productivity and environmental performance. However, the systematic identification and evaluation of high-potential digital innovation opportunities in construction remain underdeveloped, leaving firms with limited decision support at the research and development stage. To address this gap, this study adopts a technology convergence perspective and develops a two-stage machine learning framework for predicting technology opportunities. In Stage 1, the framework assesses the likelihood that a focal digital technology will converge with construction technologies and form a viable opportunity. In Stage 2, it predicts the subsequent value potential of the resulting convergence. Using 129,980 invention patents related to construction and digital technologies worldwide from 2015 to 2023, we evaluate the model’s predictive performance and further examine its robustness across technology domains, time periods, alternative high-value thresholds, and oversampling strategies. The proposed framework demonstrates strong predictive performance. In addition, Shapley additive explanations are used to interpret the model’s predictive logic, showing that knowledge-network information is the most influential predictor in both stages. This study advances understanding of innovation opportunities in construction digitalization and provides a data-driven decision-support tool for technology investment and innovation governance for firms, policymakers, and industry stakeholders.