Forecasting Construction Equipment Productivity Based on Weather Conditions: A Data-Driven Time-Series Machine Learning Approach

Accurate forecasting of construction equipment productivity is crucial for effectively planning and controlling construction resources. Traditional approaches often rely on data from previous projects and the subjective experience of project managers, which can result in inaccurate estimates. Moreover, previous studies in forecasting construction equipment productivity often tend to overlook the time dependencies in the data which is inherent to the nature of construction productivity. This study presents a generic data-driven framework for forecasting productivity data from ongoing construction field activities and weather conditions, utilizing daily report logs. The framework combines machine learning (ML) models with time-series analysis to forecast, over time, the productivity of construction equipment while considering weather conditions as exogenous factors. Modeling construction productivity as a time series enables the reflection of temporal dependencies in the data. The framework was evaluated through a case study involving excavation activities for an infrastructure project over eight months. The ML-based time-series model yielded a model fit (R2) of 0.89, outperforming the classical Seasonal Autoregressive Integrated Moving Average (SARIMAX) time-series model, which achieved an R2 of 0.71. Moreover, the ML-based time-series model showed a 63% reduction in the estimation error compared to the project manager’s approach. Once the construction field productivity is accurately estimated, project managers can leverage the estimates to make data-driven resource allocation decisions.