Change orders, defined as formal contract modifications, frequently disrupt project performance, yet predictive approaches to assess their impacts are still limited and insufficiently explainable for practical use in construction. This study introduces an explainable artificial intelligence (XAI)-based framework to profile the impact of change order-related disruptions, aiming to provide both accurate predictions and interpretable insights into associated risks. Utilizing a comprehensive dataset from highway construction projects, including project-specific variables, change order characteristics, and a macroeconomic indicator, the study develops four predictive models: two classification models to determine whether a change order causes delays or cost overruns, and two regression models to estimate the severity of these impacts. An XAI technique is then applied to identify key contributing factors and construct an interpretable impact profiling framework. Three principal findings emerge from this study. First, tree-based ensemble learning models delivered strong predictive capabilities in classifying and quantifying the impacts of change orders on schedule and cost. Second, XAI interpretation identified critical factors influencing change order-related disruptions and uncovered their value-dependent impact patterns, including monotonic and non-monotonic forms. Third, the proposed impact profiling framework, presented through a visual representation, offers dual-perspective insights by capturing both the overall importance of each feature and the value-dependent effects of features on disruption outcomes, while jointly presenting schedule- and cost-related impacts in an integrated, interpretable view. By combining predictive performance with interpretability, the proposed framework advances expert systems for intelligent and transparent decision support in construction, helping stakeholders manage change order risks and ensure resilient project delivery.
Oh et al. (2026) studied this question.