Dynamic-Bayesian-Network-Based Project Cost Overrun Prediction Model

Abstract

One common problem in the construction industry is project cost overrun. Cost overrun can have significant impacts on financial profitability, project completion, project quality, and stakeholder satisfaction. The average percentage of construction project overrun can vary widely depending on the project type, size, complexity, and location. Many approaches can be adopted to prevent or mitigate project cost overrun, and one of them is a more accurate cost estimate and prediction. Several studies on the construction project cost overrun estimation and prediction have been conducted based on historical data; nevertheless, each project has its project characteristics and cost trend. Real-time, project-specific cost data are more reliable for forecasting the cost trend of the project itself. There are many influence factors that may interdependently affect the construction project cost overrun. This paper proposes a real-time predict cost overrun risk prediction model based on the influence factors and their interdependence as well as the corrective actions if adopted. This study used a dynamic Bayesian network (DBN) to formulate problem architecture and to use the input–output hidden Markov method (I/O HMM) with particle filter (PF) to run inference. Six building and mass rapid transit projects in Taiwan were used as model validation and comparison. The posterior probabilities from the DBN-based cost overrun risk prediction model were highly consistent with the cost overrun ratios of real construction projects. Moreover, it is superior to other prediction models in terms of accuracy. The proposed model could provide project managers with an early alert for cost overrun.