Predictive Capabilities of Supervised Learning Models Compare with Time Series Models in Forecasting Construction Hiring

10 pages•Published: June 9, 2021

Abstract

The construction market is playing a massive role in the United States Gross Domestic Product (GDP). Among various industries, construction is a significant sector responsible for 4-8 percent of GDP. Like other sectors, construction markets are susceptible to demand fluctuations, which the economic recession can cause, political decisions, natural disasters, or outbursts of pandemics. The ability to predict the demand rate in the construction market could give the contractors and owners a better understanding of what they need in their short-term and long-term programs and make them more competitive by predicting the needs in workforce demand. The research selected Texas employment data as the focal point due to the size of the construction market and its workforce diversity.
Furthermore, Texas has been a hotspot for dozens of hurricanes, also affected by many political bills and economic Turmoil, making results more capable of further generalization. This research used three different methods to predict the total construction employment. Univariate models are applied to the datasets to forecast them based on their previous quantities. Three methods such as autoregressive integrated moving average time series models (ARIMA), supervised learning regressors, and the long-term-short-memories (LSTM), were applied to the construction hiring data extracted from the U.S. Bureau of Labor Statistics website. Generally, LSTM models had the most accurate predictions in most cases, except for Austin, where ARIMA models predicted the dataset accurately.

Keyphrases: construction hiring, deep learning, lstm, time series models, workforce migration

In: Tom Leathem, Anthony Perrenoud and Wesley Collins (editors). ASC 2021. 57th Annual Associated Schools of Construction International Conference, vol 2, pages 117-126.

Links:	https://easychair.org/publications/paper/wvPw
	https://doi.org/10.29007/nr1t

BibTeX entry

@inproceedings{ASC2021:Predictive_Capabilities_Supervised_Learning,
  author    = {Boong Ryoo and Milad Ashtab},
  title     = {Predictive Capabilities of Supervised Learning Models Compare with Time Series Models in Forecasting Construction Hiring},
  booktitle = {ASC 2021. 57th Annual Associated Schools of Construction International Conference},
  editor    = {Tom Leathem and Anthony Perrenoud and Wesley Collins},
  series    = {EPiC Series in Built Environment},
  volume    = {2},
  publisher = {EasyChair},
  bibsource = {EasyChair, https://easychair.org},
  issn      = {2632-881X},
  url       = {/publications/paper/wvPw},
  doi       = {10.29007/nr1t},
  pages     = {117-126},
  year      = {2021}}

Download PDF Open PDF in browser