Scholarlink

Abstract <p>Accurate prediction of unconfined compressive strength (UCS) of expansive soils is essential for geotechnical design and risk mitigation. This study investigates the application of four artificial intelligence models, including artificial neural network (ANN), support vector regression (SVR), random forest (RF), and Gaussian process regression (GPR) to predict UCS based on ten geotechnical input parameters. A dataset compiled from 145 published studies was used, encompassing a diverse range of expansive soils. All features were normalized, and model performance was evaluated using the coefficient of determination (<i>R</i><sup>2</sup>), root mean square error (RMSE), mean absolute error (MAE), bias, standard error (SE), and scatter index (SI). Among the models, ANN consistently demonstrated superior performance, achieving the highest testing <i>R</i><sup>2</sup> (0.888), lowest MAE (0.061), and lowest SI (0.297), followed closely by GPR. Error distribution and training history further confirmed the robustness and generalization capability of ANN, with well-centred, symmetric prediction errors and no signs of overfitting. The results highlight ANN as an effective predictive tool for UCS estimation, offering enhanced accuracy and reliability over conventional machine learning approaches.</p> Research highlights <p><UnorderedList Mark="Bullet"> <ItemContent> <p>Four AI models (ANN, GPR, SVR, and RF) were developed and evaluated for UCS prediction.</p> </ItemContent> <ItemContent> <p>Bayesian optimization was applied to efficiently tune model hyperparameters and enhance prediction accuracy.</p> </ItemContent> <ItemContent> <p>The ANN model achieved the best overall performance, showing superior accuracy and generalization on unseen data.</p> </ItemContent> <ItemContent> <p>Feature importance analysis revealed maximum dry density and sand content as the most influential factors controlling UCS.</p> </ItemContent> </UnorderedList></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 an artificial neural network model for predicting unconfined compressive strength of expansive soils

  • 期刊论文

  • 发表日期：

    2026年6月25日

查看全文
  • Thanh-Hai Do

摘要

Abstract

Accurate prediction of unconfined compressive strength (UCS) of expansive soils is essential for geotechnical design and risk mitigation. This study investigates the application of four artificial intelligence models, including artificial neural network (ANN), support vector regression (SVR), random forest (RF), and Gaussian process regression (GPR) to predict UCS based on ten geotechnical input parameters. A dataset compiled from 145 published studies was used, encompassing a diverse range of expansive soils. All features were normalized, and model performance was evaluated using the coefficient of determination (R2), root mean square error (RMSE), mean absolute error (MAE), bias, standard error (SE), and scatter index (SI). Among the models, ANN consistently demonstrated superior performance, achieving the highest testing R2 (0.888), lowest MAE (0.061), and lowest SI (0.297), followed closely by GPR. Error distribution and training history further confirmed the robustness and generalization capability of ANN, with well-centred, symmetric prediction errors and no signs of overfitting. The results highlight ANN as an effective predictive tool for UCS estimation, offering enhanced accuracy and reliability over conventional machine learning approaches.

Research highlights

Four AI models (ANN, GPR, SVR, and RF) were developed and evaluated for UCS prediction.

Bayesian optimization was applied to efficiently tune model hyperparameters and enhance prediction accuracy.

The ANN model achieved the best overall performance, showing superior accuracy and generalization on unseen data.

Feature importance analysis revealed maximum dry density and sand content as the most influential factors controlling UCS.