Machine Learning–Enhanced Recurrent Event Modeling for Change Order Recurrence in Highway Construction

Abstract

Change orders are a persistent challenge in construction projects, often resulting in substantial schedule delays and budget overruns. This study examined the recurrence of change orders by analyzing 1,182 change orders across 68 highway construction projects. The objectives of the research were threefold: (1) to identify key factors influencing change order recurrence; (2) to assess how recurrence patterns evolve throughout the project lifecycle; and (3) to evaluate their impact on project outcomes at different stages. A hybrid analytical approach integrating recurrent event modeling and machine learning (ML), along with statistical tests, was utilized to achieve these goals. The findings highlight significant predictors of change order recurrence, including time-dependent factors (e.g., original contract amount), time-independent factors (e.g., change in contract duration), and an ML-derived risk score. This study finds that larger-scale projects, the low-bid-procured design-bid-build delivery method, higher contingency levels, minor contract duration extensions, and Fall-season change orders are linked to increased recurrence, particularly in the early stage of a project. In contrast, as the project progresses, the effects of recurrent change orders on schedules and costs become more pronounced. Based on these insights, this study proposed phase-specific and cross-phase strategies to mitigate the risks associated with recurrent change orders. Through advanced analytical techniques, this research contributes to the body of knowledge in risk management and project planning, offering a robust framework for understanding change order recurrence.