Steel, copper, and aluminum are essential raw materials in the construction industry and are used in any construction project whether commercial, industrial, or mega projects. During normal market conditions, the prices of these metals normally fluctuate due to general inflation in the economy; however, during crises, the prices of these metals significantly fluctuate exposing both contractors and suppliers to an unexpected financial risk. Various researchers have developed metal prediction models for normal market condition and did not consider testing it on the crises time; also, the developed models are devised using past metal prices or macroeconomic indicators of trading countries and limited research is found in using other external parameters of the exporting countries that can indicate the trend of these metal prices. Thereto, this research aims to develop two general price prediction models for each metal that can forecast the monthly metal prices of steel rebar traded in the Shanghai Metal Exchange, copper and aluminum traded in the London Metal Exchange, for both the standard and crises times. The first is a multivariate model that can predict the metal prices for 1 month in the future using external parameters and the second is a univariate model that can forecast the metal prices for the upcoming 3 months or more using the past metal prices. For the purpose of developing generic price prediction models, the time frame considered in this research is January 2009 till July 2022 and the models developed are built using historical data and are validated and tested on the recent crises, COVID-19 and Russia-Ukraine war. The tool adopted is the Long Short-Term Memory network (LSTM) using the Python programming language in jupyter notebook. Some potential input variables are considered and used as input variables to the multivariate models. This includes the metal raw material prices, the inflation rate and international reserve of USD and assets of major exporting countries of metal, the transportation index, and a global market indicator. So, prior to developing the LSTM models, the collected input variables are analyzed with respect to the output variable using correlation, granger-causality and multicollinearity tests in R software; then, the input variables are further filtered to select the significantly statistical input variables that impact the output variable. Then, the LSTM models are developed for each metal, and the dataset is divided into training, validation, and testing sets and mean square errors, root mean square errors, and absolute percentage errors are used to assess the model’s performance. The developed models show proposing results where the absolute percentage error ranges between 0.1% and 18.5% with a mean absolute percentage error below 10% for all the developed models. Also, it is concluded that the performance of the 3-Month univariate prediction model is better than the 1-Month multivariate prediction model for all the metals, but still, both models can act as proposing tools for decision-making times of high price fluctuation or even in regular price fluctuation times.


School of Sciences and Engineering


Construction Engineering Department

Degree Name

MS in Construction Engineering

Graduation Date

Winter 1-31-2023

Submission Date


First Advisor

Dr. Ahmed Samer Ezeldin

Second Advisor

Dr. Engy Serag

Committee Member 1

Dr. Mohamed Mahdy

Committee Member 2

Dr. Ossama Hosny

Committee Member 3

Dr. Ibrahim Abotaleb


192 p.

Document Type

Master's Thesis

Institutional Review Board (IRB) Approval

Not necessary for this item

Available for download on Thursday, January 23, 2025