Data-Driven Schedule Risk Forecasting for Construction Mega-Projects

Abstract

Accurately forecasting and mitigating schedule risks in construction projects is an incredibly valuable and equally challenging task. In recent years this task has gained added attention from the machine learning community. State-of-the-art methods, however, both in academia and in industry still rely on expert opinions and heuristic methods for estimating parameterized models. This paper studies the performance of machine learning models compared to more traditional state-of-the-art approaches for construction mega-project schedule risk forecasting. To better understand the importance of data-driven methods for project risk forecasting, extensive experimental results on thousands of projects from various industries and sectors are reported. These results convey a clear message: construction mega-project schedule risks should be analyzed using data-driven models to enable more accurate and scalable risk analyses when appropriate data is available. Based on these observations an outlook for further developments in academia and industry both from the machine learning and project risk management perspectives is suggested.

Keywords: construction, scheduling, machine learning, graph neural networks, pert