Forecasting Pipeline Construction Costs Using Recurrent Neural Networks

Abstract

Pipe material and labor costs comprise about 70% of the total pipeline construction cost. Pipe and labor costs experience volatile fluctuations over time, which mostly cause cost overruns in lengthy and large-scale pipeline projects. The accurate forecasting of pipe and labor costs is critical for cost estimators to prepare accurate bids and manage the cost contingencies. The objective of this research is to develop recurrent neural networks (RNNs) to forecast pipeline construction costs. Pipe material (reinforced concrete pipe, corrugated steel pipe) and labor (common labor, skilled labor) costs from January 1995 to December 2018 were collected from Engineering News-Record (ENR) to develop the RNNs. The out-of-sample forecasting accuracies of the RNNs were validated using the ENR pipe and labor cost observations of 12 months in 2019. The results show that the RNNs consistently outperform the seasonal autoregressive integrated moving average (SARIMA) models, which are the most accurate univariate time series model in forecasting pipe and labor cost fluctuations. This research contributes to the pipeline construction community by assisting cost engineers and project managers in enhancing bidding, budgeting, and cost estimating for pipeline projects.