Chapter 2: A Machine Learning–Based Buffer Estimation for Construction Schedules Under Uncertainty

Uncertainty in construction projects is a global phenomenon that causes delays and incurs extra costs throughout the execution phase. Delays can be mitigated and tolerated if predicting the total duration in the planning phase involves uncertainty factors and is based on data-driven model estimation instead of experience-based estimations only. This chapter introduces a buffer allocation proactive scheduling approach based on machine learning (ML) predictions. The developed ML models (Regression) will predict the as-built durations considering a degree of uncertainty for the project activities during the planning phase; these estimates will be used to create a proactive baseline schedule that is able to tolerate uncertainties throughout the execution phase. During the execution phase of the projects, some unexpected events and uncertainties might occur; henceforth, this proactive scheduling approach is developed to control and mitigate these events when occurring. A comparison between several ML models for the prediction of the final duration and of construction projects is done, and based on the performance measures, Gaussian Process Regression outperformed the other algorithms. Five parameters will be used for the ML regression model (As-planned duration, Contractor, Weather Conditions, Owner, Consultant) that were expressed through field interviews as highly effective delay factors in the construction projects, in addition to a review of the existing literature. Historical data of previously executed projects will be collected to train the models for prediction based on cross-validation. A complete data sample of 78 construction projects was collected from the construction industry in Jordan, with a total of 122 Primavera (P6) as-planned and as-built schedules.