Scholarlink

Abstract <p>The California Bearing Ratio (CBR) is critical parameter for evaluating geotechnical projects to ensure soil strength and its bearing capacity for roadway applications. This study highlights advanced predictive models including Full Quadratic (FQ), Interaction (IA), Artificial Neural Network (ANN), and Random Forest (RF) to evaluate the CBR of Rice Husk Ash (RHA)-treated and untreated soils. Also, no studies have scientifically compared multiple advanced models, such as FQ, IA, ANN, and RF for predicting the CBR values of RHA-stabilized soils. Input variables such as RHA%, liquid limit (LL), plasticity index (PI), maximum dry density (MDD), optimum moisture content (OMC), and CBR (soaked and unsoaked) conditions were considered and assessed on 166 dataset samples. According to the parameter metrics, such as R<sup>2</sup>, RMSE, MAE, and SI, the model outcomes demonstrated that ANN model obtained the highest accuracy with (R<sup>2</sup> = 0.97, RMSE = 0.85%, MAE = 0.61%, SI = 0.007), for both soaked, and unsoaked conditions. Moreover, full quadratic and interaction models indicated practical performance but were less impact in capturing non-linear relationships. Sensitivity analysis revealed that CBR conditions, and RHA were the most impact factors, with OMC and LL also playing significant roles. The study showed an optimal range of RHA content for best CBR improvement, and supporting the sustainable use of RHA as a soil stabilizer. Based on the results, the models proved high accuracy, reducing time-intensive laboratory tests while offering reliable predictions for soil stabilization projects.</p> Graphical Abstract <p></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

Comparative Assessment of Advanced Machine Learning Models for Predict the California Bearing Ratio of Rice Husk Ash – Stabilized Soils

  • 期刊论文

  • 发表日期：

    2026年2月11日

查看全文
  • Rizgar A. Blayi,
  • Jamal I. Kakrasul,
  • Samir M. Hamad

摘要

Abstract

The California Bearing Ratio (CBR) is critical parameter for evaluating geotechnical projects to ensure soil strength and its bearing capacity for roadway applications. This study highlights advanced predictive models including Full Quadratic (FQ), Interaction (IA), Artificial Neural Network (ANN), and Random Forest (RF) to evaluate the CBR of Rice Husk Ash (RHA)-treated and untreated soils. Also, no studies have scientifically compared multiple advanced models, such as FQ, IA, ANN, and RF for predicting the CBR values of RHA-stabilized soils. Input variables such as RHA%, liquid limit (LL), plasticity index (PI), maximum dry density (MDD), optimum moisture content (OMC), and CBR (soaked and unsoaked) conditions were considered and assessed on 166 dataset samples. According to the parameter metrics, such as R2, RMSE, MAE, and SI, the model outcomes demonstrated that ANN model obtained the highest accuracy with (R2 = 0.97, RMSE = 0.85%, MAE = 0.61%, SI = 0.007), for both soaked, and unsoaked conditions. Moreover, full quadratic and interaction models indicated practical performance but were less impact in capturing non-linear relationships. Sensitivity analysis revealed that CBR conditions, and RHA were the most impact factors, with OMC and LL also playing significant roles. The study showed an optimal range of RHA content for best CBR improvement, and supporting the sustainable use of RHA as a soil stabilizer. Based on the results, the models proved high accuracy, reducing time-intensive laboratory tests while offering reliable predictions for soil stabilization projects.

Graphical Abstract