Scholarlink

<p>The free swell index (FSI) is an important parameter for characterising expansive soils, which pose significant geotechnical challenges. This study focuses on classifying the expansive nature of soils and predicting the FSI using easily determined and experimentally measured soil properties by employing various classification and regression machine learning algorithms. A total of 100 soil samples were experimentally tested and analysed to determine clay content, liquid limit, plastic limit, plasticity index, activity, and free swell index. Classification models (decision tree, random forest, k-nearest neighbor and naïve bayes) were employed to distinguish between expansive and non-expansive soils, while regression models (random forest regression, gradient boosting regression, categorical boosting regression, artificial neural network and bootstrapped multiple linear regression) were used to predict the FSI. Twenty-seven models were developed across three training-to-testing ratios: 70:30, 75:25 and 80:20. The k-nearest neighbor (k-NN) model was identified as the optimum performance model for classifying expansive soils with ideal values (1.00) for accuracy, precision, recall and AUC. Furthermore, ideal scores for TNR, NPV, FOR, ACA and MCC highlighted the robustness of the k-NN model in distinguishing between expansive and non-expansive soils. The artificial neural network (ANN) model was recognized as the optimum performance model for predicting the FSI with an R<sup>2</sup> of 0.82, RMSE of 6.13, MAE of 5.36, RSR of 0.45, NMBE of 0.149, PI of 4.14, a20 of 0.171, and IOA of 0.934. These findings demonstrate the potential of data-driven approaches to effectively classify expansive soils and predict their swelling behaviour, thus improving early-stage project planning.</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

Prediction of soil swelling using machine learning techniques

  • 期刊论文

  • 发表日期：

    2025年8月19日

查看全文
  • Zaffir Mohammed,
  • Gregory Gouveia,
  • Amit Ramkissoon,
  • Ricardo Clarke

摘要

The free swell index (FSI) is an important parameter for characterising expansive soils, which pose significant geotechnical challenges. This study focuses on classifying the expansive nature of soils and predicting the FSI using easily determined and experimentally measured soil properties by employing various classification and regression machine learning algorithms. A total of 100 soil samples were experimentally tested and analysed to determine clay content, liquid limit, plastic limit, plasticity index, activity, and free swell index. Classification models (decision tree, random forest, k-nearest neighbor and naïve bayes) were employed to distinguish between expansive and non-expansive soils, while regression models (random forest regression, gradient boosting regression, categorical boosting regression, artificial neural network and bootstrapped multiple linear regression) were used to predict the FSI. Twenty-seven models were developed across three training-to-testing ratios: 70:30, 75:25 and 80:20. The k-nearest neighbor (k-NN) model was identified as the optimum performance model for classifying expansive soils with ideal values (1.00) for accuracy, precision, recall and AUC. Furthermore, ideal scores for TNR, NPV, FOR, ACA and MCC highlighted the robustness of the k-NN model in distinguishing between expansive and non-expansive soils. The artificial neural network (ANN) model was recognized as the optimum performance model for predicting the FSI with an R2 of 0.82, RMSE of 6.13, MAE of 5.36, RSR of 0.45, NMBE of 0.149, PI of 4.14, a20 of 0.171, and IOA of 0.934. These findings demonstrate the potential of data-driven approaches to effectively classify expansive soils and predict their swelling behaviour, thus improving early-stage project planning.