Scholarlink

<p>Iron ore tailings are by-products formed as a result of iron ore beneficiation and extraction processes. 1.4&#xa0;billion tons of tailings are produced annually, and Australia, Brazil, and China are the major producers. Magnetite tailings are used to improve soil strength parameters. Still, the repeated load triaxial test (AASHTO T307) for determining the resilience modulus of treated soil is lengthy and requires human resources. This study employs an extreme gradient boosting (XGBoost) model, optimized by Arithmetic Optimization (AOA), Brainstorm Optimization (BOA), Tiki-Taka Optimization (TTA), and Whale Optimization (WOA) algorithms, and compares them to introduce an optimal performance model to predict the resilience modulus (M<sub>R</sub>) of cement-treated magnetite iron ore tailings (MIOT). A database of 210 M<sub>R</sub> of cement-treated MIOT specimens was used to develop and analyze XGBoost models. The performance metrices such as root mean square error (RMSE), variance accounted for (VAF), performance index (PI), and bias factor (BF) demonstrated that XGBoost_TTA (RMSE = 20.5498&#xa0;MPa, VAF = 99.07, PI = 1.96, and BF = 0.9873) outperformed the XGBoost, XGBoost_AOA, XGBoost_BOA, and XGBoost_WOA models because the TTA algorithm maintains a balance between exploration and exploitation, and helps prevent the model from getting stuck in local optima. The Taylor and QQ plots visually illustrated the XGBoost_TTA model’s ability to predict M<sub>R</sub>. The rank, regression error characteristics (REC), accuracy matrix, and Akaike information criterion (AIC) analyses revealed the robustness of the XGBoost_TTA model over the conventional and optimized XGBoost models. In addition, the overfitting and generalizability of each model are evaluated and analyzed. Shapley Additive Explanations (SHAP) analysis identified bulk stress as the dominant predictor (mean |SHAP| = 126.01), followed by cement content (70.09), deviatoric stress (34.13), and curing time (25.28), with partial dependence plots revealing non-linear relationships for cement content and bulk stress, linear behavior for curing time, and saturation effects for deviatoric stress.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Application of Hybrid Extreme Gradient Boosting Models for Predicting the Resilient Modulus of Cement-Stabilized Magnetite Iron Ore Tailings

  • 期刊论文

  • 发表日期：

    2025年11月19日

查看全文
  • Avinash Kumar,
  • Aradhana Kumari,
  • Sudeep Kumar,
  • Jitendra Khatti,
  • Abidhan Bardhan

摘要

Iron ore tailings are by-products formed as a result of iron ore beneficiation and extraction processes. 1.4 billion tons of tailings are produced annually, and Australia, Brazil, and China are the major producers. Magnetite tailings are used to improve soil strength parameters. Still, the repeated load triaxial test (AASHTO T307) for determining the resilience modulus of treated soil is lengthy and requires human resources. This study employs an extreme gradient boosting (XGBoost) model, optimized by Arithmetic Optimization (AOA), Brainstorm Optimization (BOA), Tiki-Taka Optimization (TTA), and Whale Optimization (WOA) algorithms, and compares them to introduce an optimal performance model to predict the resilience modulus (MR) of cement-treated magnetite iron ore tailings (MIOT). A database of 210 MR of cement-treated MIOT specimens was used to develop and analyze XGBoost models. The performance metrices such as root mean square error (RMSE), variance accounted for (VAF), performance index (PI), and bias factor (BF) demonstrated that XGBoost_TTA (RMSE = 20.5498 MPa, VAF = 99.07, PI = 1.96, and BF = 0.9873) outperformed the XGBoost, XGBoost_AOA, XGBoost_BOA, and XGBoost_WOA models because the TTA algorithm maintains a balance between exploration and exploitation, and helps prevent the model from getting stuck in local optima. The Taylor and QQ plots visually illustrated the XGBoost_TTA model’s ability to predict MR. The rank, regression error characteristics (REC), accuracy matrix, and Akaike information criterion (AIC) analyses revealed the robustness of the XGBoost_TTA model over the conventional and optimized XGBoost models. In addition, the overfitting and generalizability of each model are evaluated and analyzed. Shapley Additive Explanations (SHAP) analysis identified bulk stress as the dominant predictor (mean |SHAP| = 126.01), followed by cement content (70.09), deviatoric stress (34.13), and curing time (25.28), with partial dependence plots revealing non-linear relationships for cement content and bulk stress, linear behavior for curing time, and saturation effects for deviatoric stress.