نوع مقاله : مقاله پژوهشی
نویسندگان
1 دانشیاردانشکده مهندسی معدن، دانشگاه صنعتی امیرکبیر
2 دانشکده مهندسی معدن، دانشگاه صنعتی امیرکبیر
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
The cut-off grade is a vital parameter in mining economics, investment decisions, and feasibility studies of mining projects, with mineral commodity prices playing a key role in its optimization—especially in open-pit metal mines. Consequently, forecasting iron ore prices using Artificial Intelligent methods offers valuable insights for mine design and planning. In this study, monthly data for 14 variables—including iron ore price, base metals (copper, aluminum, lead, zinc, tin), precious metals (gold, silver), energy carriers (coal, liquid fuel, natural gas), and economic indicators (global iron production and demand, average transportation cost)—were collected and analyzed using Bivariate Linear Regression to identify relationships between iron ore price and each variable. Strong correlations were found with copper, lead, gold, silver, tin, natural gas, liquid fuel, global iron production and demand, and transportation costs, indicating their significant influence on iron ore pricing. After normalizing and splitting the data into training and testing sets, Machine Learning models were implemented—including Multiple Linear Regression, Support Vector Regression, Decision Tree, Random Forest, and eXtreme Gradient Boosting (XGBoost)—and evaluated using R², MAPE, MSE, and MAE. Tree-based models revealed that precious metal prices had the highest feature importance, while energy prices had the lowest. Among all models, XGBoost demonstrated superior performance with an R² of 0.93 and MAE of 0.15, effectively capturing complex price fluctuations and outperforming traditional regression approaches. This model provides a reliable and powerful foundation for forecasting iron ore prices, supporting strategic planning and optimization in the mining industry. representation of the correlation matrix helps identify the features most strongly associated with the dependent variable. As a result, ten parameters—including the prices of copper, lead, gold, silver, tin, gas, LNG, global iron ore production and demand, and transportation costs—were chosen as critical features for modelling. According to results of the constructed prediction models ,, the XGB model outperformed the best performance and accuracy, while MLR model had the weakest results. the best-fit ML model in this investigation is XGB for world iron ore price prediction based on higher performance as it gave the highest R2 (0.93) and lowest MAE (0.15) compared to other models.
کلیدواژهها [English]
)