Forecasting Regional Disparities in Construction Labor Earnings Using Long Short-Term Memory (LSTM)

Skilled labor shortages, rapid increases in demand for construction activities, and changes to the labor union working environments create volatile shifts in labor costs. Although existing studies used national-level data to capture the volatility in labor costs, the state-level disparities in local economic conditions, and composition of the workforce, render national-level estimates suboptimal. The goal of this paper is to enhance the management of construction labor costs in state-level markets. To this end, univariate and multivariate long short-term memory (LSTM) models were developed to predict labor earnings. Results reveal that the multivariate LSTM models predict the state-level labor earnings with an accuracy exceeding 97%. Analysis of the residuals of the LSTM models supports their fidelity. Accurate prediction of state-level labor earnings provides construction contractors and project owners with a valuable tool to prepare accurate labor cost estimates, manage resource allocation during execution, and proactively enhance budgeting decisions in different markets.