Scholarlink

<p>Accurate prediction of the California Bearing Ratio (CBR) is crucial for optimizing the assessment of subgrade soils in pavement design, especially in tropical regions where lateritic gravels are widely used. Conventional CBR testing is time-consuming, costly, and labor-intensive. This study evaluates the use of 03 machine-learning algorithms namely, K-Nearest Neighbors (KNN), Decision Tree Regressor (DTR), and Support Vector Machine (SVM) to estimate CBR based on 14 geotechnical parameters, including 04 derived indices: grading modulus (Gm), plasticity modulus (Pm), plasticity product (Pp), and swelling potential (εs). A dataset of 210 samples collected in East Cameroon were analyzed in accordance with AFNOR (Association Française de Normalisation) standards. To improve model robustness, 5-fold cross-validation and data normalization were applied. Among the tested models, DTR achieved the highest predictive performance (coefficient of determination, R² = 0.97), followed by SVM (R² = 0.92) and KNN (R² = 0.86). Paired t-tests confirmed that the differences in model performance are statistically significant at the 1% level within the present dataset and study conditions. Incorporating derived parameters improved performance by 14% for KNN, 162% for DTR, and 441% for SVM. Beyond the local context, the proposed framework provides a transferable methodology for predicting the mechanical performance of lateritic gravels across tropical regions worldwide, offering a rapid, cost-effective, and reliable complement to traditional CBR testing.</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

Assessment of CBR of granite-derived lateritic gravels using machine-learning models with derived parameters

  • 期刊论文

  • 发表日期：

    2026年4月8日

查看全文
  • Willy Passy Mvindi Mvindi,
  • Aloys Thierry Ndzié Mvindi,
  • Hervé Nyassa Ohandja,
  • Ghislain Nguentchou Tchanfa,
  • Danielle Léa Mebouinz,
  • Sylvie Désirée Noa Tang,
  • Vincent Laurent Onana

摘要

Accurate prediction of the California Bearing Ratio (CBR) is crucial for optimizing the assessment of subgrade soils in pavement design, especially in tropical regions where lateritic gravels are widely used. Conventional CBR testing is time-consuming, costly, and labor-intensive. This study evaluates the use of 03 machine-learning algorithms namely, K-Nearest Neighbors (KNN), Decision Tree Regressor (DTR), and Support Vector Machine (SVM) to estimate CBR based on 14 geotechnical parameters, including 04 derived indices: grading modulus (Gm), plasticity modulus (Pm), plasticity product (Pp), and swelling potential (εs). A dataset of 210 samples collected in East Cameroon were analyzed in accordance with AFNOR (Association Française de Normalisation) standards. To improve model robustness, 5-fold cross-validation and data normalization were applied. Among the tested models, DTR achieved the highest predictive performance (coefficient of determination, R² = 0.97), followed by SVM (R² = 0.92) and KNN (R² = 0.86). Paired t-tests confirmed that the differences in model performance are statistically significant at the 1% level within the present dataset and study conditions. Incorporating derived parameters improved performance by 14% for KNN, 162% for DTR, and 441% for SVM. Beyond the local context, the proposed framework provides a transferable methodology for predicting the mechanical performance of lateritic gravels across tropical regions worldwide, offering a rapid, cost-effective, and reliable complement to traditional CBR testing.