Scholarlink

<p>This research provides an investigation of soaked California Bearing Ratio (CBR) values derived from a dataset of 200 soil samples, utilizing various Support Vector Machine (SVM) models. Nine SVM configurations were assessed, encompassing Linear, Quadratic, Cubic, Fine, Medium, Coarse, and Cubic models utilizing Gaussian, Linear, and Quadratic kernels. The Cubic SVM with Quadratic Kernel Function exhibited superior predictive performance, attaining the lowest testing Root Mean Square Error (RMSE) of 0.701, Mean Squared Error (MSE) of 0.491, and the highest R² of 0.916. The Cubic SVM utilizing a Gaussian Kernel demonstrated strong performance, achieving a testing RMSE of 0.865 and a R² of 0.872. In contrast, the Linear SVM exhibited a lower R² of 0.582 and a higher RMSE of 1.563, attributable to its limitations in modeling non-linear relationships. Interpretability analyses utilizing SHapley Additive exPlanations (SHAP), local interpretable model-agnostic explanations (LIME), and Partial Dependence Plots (PDPs) revealed that % Sand, % Clay, and Liquid Limit are the most significant features. The advanced kernel models successfully captured the complex non-linear relationships within the data, as evidenced by the smooth response plot trends and consistent feature contributions. Conversely, simpler models such as Linear SVM demonstrated constraints in both interpretability and predictive accuracy. This research offers reliable, interpretable, and accurate predictions of soil CBR values, which are crucial for foundation and pavement design.</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

Explainable AI Based Support Vector Machine Models for Soaked CBR Prediction

  • 期刊论文

  • 发表日期：

    2025年5月23日

查看全文
  • M. K. Sachin Gowda,
  • Yogesh Iyer Murthy,
  • Aakash Gupta

摘要

This research provides an investigation of soaked California Bearing Ratio (CBR) values derived from a dataset of 200 soil samples, utilizing various Support Vector Machine (SVM) models. Nine SVM configurations were assessed, encompassing Linear, Quadratic, Cubic, Fine, Medium, Coarse, and Cubic models utilizing Gaussian, Linear, and Quadratic kernels. The Cubic SVM with Quadratic Kernel Function exhibited superior predictive performance, attaining the lowest testing Root Mean Square Error (RMSE) of 0.701, Mean Squared Error (MSE) of 0.491, and the highest R² of 0.916. The Cubic SVM utilizing a Gaussian Kernel demonstrated strong performance, achieving a testing RMSE of 0.865 and a R² of 0.872. In contrast, the Linear SVM exhibited a lower R² of 0.582 and a higher RMSE of 1.563, attributable to its limitations in modeling non-linear relationships. Interpretability analyses utilizing SHapley Additive exPlanations (SHAP), local interpretable model-agnostic explanations (LIME), and Partial Dependence Plots (PDPs) revealed that % Sand, % Clay, and Liquid Limit are the most significant features. The advanced kernel models successfully captured the complex non-linear relationships within the data, as evidenced by the smooth response plot trends and consistent feature contributions. Conversely, simpler models such as Linear SVM demonstrated constraints in both interpretability and predictive accuracy. This research offers reliable, interpretable, and accurate predictions of soil CBR values, which are crucial for foundation and pavement design.