Abstract

Change orders are formal modifications to the scope of construction projects and play a critical role in shaping project outcomes. However, predicting change orders early in a project's lifecycle remains challenging due to their complex nature and the diverse factors influencing their occurrence. These include project specifications, spatial properties, and the performance of involved actors. Despite extensive research in construction change order management, predictive models for change orders have received limited attention.
This study aims to address this gap by exploring the correlation between project attributes and change order occurrences. It focuses on abstract and easily accessible project attributes such as project type and spatial features, which are more readily shared across the industry. A new set of attributes, derived from domain knowledge, is introduced to quantify project-specific change performance and enhance the prediction of change severity.
The study also tackles the challenge of change timing prediction by modeling the temporal dependence of change orders. Using a Markov Chain approach, it simplifies the relationship between change orders issued in different project phases, assuming each phase’s outcome depends solely on the preceding phase. The validity of this assumption is tested through the Chapman-Kolmogorov equation across projects of varying durations.
Results demonstrate a 15% improvement in change severity prediction performance, highlighting the effectiveness of the introduced attributes and feature selection techniques. The findings also confirm the suitability of the Markov Chain model for capturing temporal dependencies in change orders’ severity, offering valuable insights for early change prediction and better project planning.