A Comparative Analysis of Machine Learning Approaches for Predicting Construction Steel Prices

Abstract

The volatility of steel pricing constrains the construction sector, which heavily depends on this essential material. Consequently, the development of accurate and reliable forecasting models is crucial for predicting steel prices, facilitating informed decision making, and enabling strategic planning. This study assesses ten regression, machine learning, and deep learning algorithms for predicting construction steel prices. A dataset including 7,074 historical observations was assembled, encompassing variables such as the consumer price index, foreign exchange rate, inflation rate, interest rate, unemployment rate, foreign reserves, and lending rate. The efficacy of the created models was evaluated utilizing various measures, including mean squared error (MSE), mean absolute error (MAE), root mean squared error (RMSE), and R-squared. The findings demonstrate that novel ensemble algorithms surpass single-estimator methods. The MSE of Extra Trees and Random Forest models is significantly lower than that of classic predictive models. Furthermore, the Long Short-Term Memory and Gated Recurrent Unit algorithms effectively capture temporal relationships, yielding improved performance in complicated time-series prediction tasks. The results emphasize the comparative accuracy and reliability of the examined modeling strategies for forecasting construction steel prices, assisting decision makers in selecting predictive modeling approaches.