Simulation-Based Framework for Predicting Construction Workforce Demand: A Comparative Analysis with Multivariate LSTM-Based Seq2Seq Model

Abstract

The US construction industry is currently facing a significant labor demand, making it crucial to anticipate future gaps between workforce demand and supply to enable effective planning. To address this challenge, this paper proposes a simulation-based framework for estimating and predicting future workforce needs. The framework’s applicability and effectiveness are demonstrated through two case studies of the Austin–Round Rock metropolitan statistical area (MSA) and the Dallas–Fort Worth–Arlington MSA. Additionally, a multivariate long short-term memory (LSTM) encoder–decoder-based sequence-to-sequence (Seq2Seq) model is developed for each MSA to serve as a statistical model for comparison. The performance of the developed agent-based modeling approach is then compared with the Seq2Seq model. The case study results suggest that the simulation model outperforms the statistical model in the face of unexpected events such as Covid-19 outbreaks with lower mean absolute percentage error values of 1.34% and 0.88% for the Austin–Round Rock MSA and the Dallas–Fort Worth–Arlington MSA, respectively. The proposed model offers a valuable tool for industry practitioners seeking to accurately estimate and predict future workforce demand and supply in the construction industry.