A Prediction of Cement and Iron Product Price Index Based on Machine Learning Algorithm by Using Extreme Gradient Boosting (XGBoost)

  • วิชญ์พล สุดสงวน ภาควิชาวิศวกรรมโยธา คณะวิศวกรรมศาสตร์ จุฬาลงกรณ์มหาวิทยาลัย จ.กรุงเทพมหานคร
  • วัชระ เพียรสุภาพ ภาควิชาวิศวกรรมโยธา คณะวิศวกรรมศาสตร์ จุฬาลงกรณ์มหาวิทยาลัย จ.กรุงเทพมหานคร


Construction cost significantly impacts the analysis of feasibility, budget planning, and project success. For contractors, construction cost can be used to determine the bid price and profit. Construction Material Price Index (CMI) is a useful indicator for estimating costs and planning projects, as changes in material price can impact a contractor's ability to control construction costs. This research aims to apply an Extreme Gradient Boosting (XGBoost) for developing time series forecasting model of construction material prices in Thailand. The scope of this paper is focused on cement and iron products price index. The study collected eight influencing factors over a 276-month period from January 2000 to December 2022. The data was divided into three sections: model training, model validation, and model testing. The study evaluated the model using the walk-forward optimization technique. The evaluation of forecasting accuracy was done using Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE). The research results showed that the XGBoost models with multivariate and rolling window technique outperformed the univariate models and models without the rolling window technique in terms of RMSE and MAPE on both material price index. The developed models in this study offer an approach for forecasting construction material price index, providing an accurate short-term forecast of the material price index. The developed models can serve as a tool for stakeholders in the construction industry to forecast construction material price index.