Path Analysis of Digital Twin Functions for Carbon Reduction in the Construction Industry in Hebei Province, China: A PLS-SEM and Machine Learning Approach

Jiachen Sun, Atasya Osmadi, Shan Liu, Hengbing Yin

As a significant source of global carbon emissions, the construction industry (CI) urgently needs to promote green transformation with the help of digital twin (DT) against the backdrop of human–machine collaboration and sustainable development advocated by CI 5.0. However, there is still a lack of systematic research on its specific driving mechanism and carbon reduction path. This study uses a systematic literature review (SLR) to explore how five key DT-enabled capabilities, namely, resource management (RM), process optimization (PO), real-time monitoring (R-Tm), sustainable design (SD), and predictive maintenance (PM), influence three performance indicators: efficiency improvement (EI), energy optimization (EO), and cost control (CC). Data from 490 companies were analyzed using partial least squares structural equation modeling (PLS-SEM) and a multilayer perceptron (MLP) with Shapley additive explanation (SHAP). The results show that the PLS-SEM and MLP models showed consistent patterns, with EO exhibiting the strongest predictive performance (Q2 = 0.372; R2 = 0.3666), followed by EI (Q2 = 0.307; R2 = 0.3109) and CC (Q2 = 0.305; R2 = 0.2609); the SHAP results further indicated that RM contributed most to EI (0.242), while PO was the most important driver for both EO (0.304) and CC (0.259). Academically, it introduces a quantitative approach combining PLS-SEM and machine learning. Practically, it highlights the priority of key technologies with cross-dimensional effects and offers guidance for governments to optimize digital resource allocation and carbon performance evaluation, as well as for enterprises to apply DT more effectively.