A prediction of construction material price index (CMI) by using extreme gradient boosting (XGBoost)

Abstract

This research aims to develop a time series forecasting model for construction material price index (CMI) using the Extreme Gradient Boosting (XGBoost). The study focuses on forecasting the CMI changes for 1, 3, 6, 9, and 12 months in advance. The research methodology begins with collecting eight factors that affect the forecasting, including the CMI, producer price index, consumer price index, inflation rate, interest rate, exchange rate, gross domestic product, and crude oil price, over a 276-month period from January 2000 to December 2022. Development of a time series forecasting model can be divided into four formats, namely: a univariate model, a multivariate model incorporating macroeconomic factors, a univariate model with the rolling window technique, and a multivariate model with the rolling window technique. The forecasting accuracy is evaluated using Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE). Moreover, this research also developed the Artificial Neural Networks (ANN) model in order to provide a comprehensive comparison with the XGBoost model. The results from the comparison of all four forecasting models demonstrate that the multivariate models with the rolling window technique outperform the other models in terms of RMSE and MAPE. The sanitary ware price index exhibiting the most favorable results with the RMSE ranges from 0.150 to 0.945 and the MAPE ranges from 0.129% to 0.928%, while the iron products price index displayed the highest RMSE and MAPE values with the RMSE ranges from 5.021 to 8.160, and the MAPE ranges from 2.914% to 4.602%. Furthermore, a comparison with the Artificial Neural Networks (ANN) models reveals that the XGBoost model consistently outperforms the ANN model in terms of accuracy for most construction material price index. The CMI obtained from the forecasting model can be used to support preliminary construction cost estimating and assist government agencies in calculating budgets for compensating construction costs due to fluctuations in construction material prices.