Improving decision-making in construction: Nonparametric modeling of weather-induced impacts

Effective management of construction projects encounters significant challenges, particularly due to frequent impacts influenced by climatic variables. Anticipating weather-induced impacts is crucial, and while various methods such as stochastic generators, productivity impact models, and machine learning tools are employed to model such gap, there remains a notable gap. Specifically, there is a lack of direct approaches that utilize easily accessible historical meteorological data to model impacts. This paper proposes a flexible non-parametric model designed to estimate the probability distribution of construction project completion times by incorporating meteorological information. The proposed model enhances PERT-based approaches by incorporating climatic effects to improve delay distribution estimation. This enhancement involves the direct modeling of productivity, coupled with the use of Monte Carlo simulation. A commercial version of the algorithm presented is currently part of a solution utilized in over 70 projects across the UK. The results demonstrate the model’s utility in forecasting optimal start dates, estimating project completion times, establishing contractual limits for expected weather-related delays, and analyzing critical paths within the project. Additionally, a mathematically rigorous tool for model comparison is presented, allowing for hyperparameter optimization and the selection of the most suitable prediction model.