<p>This study presents a comprehensive exploration of the development of diverse stacking machine learning models for predicting the&#xa0;Soil Liquid Index (LI)&#xa0;and&#xa0;Soil Undrained Shear Strength (S<sub>u</sub>). The study utilizes a wide range of datasets, including over 1550 data points for LI and 950 for Su, collected from various research studies. Individual models such as&#xa0;Multilayer Perceptron (MLP),&#xa0;Random Forest (RF), and&#xa0;AdaBoost regressors&#xa0;were evaluated alongside ensemble models, including&#xa0;Simple Averaging (SA),&#xa0;Weighted Averaging (WA), and&#xa0;stacking-SVR. The optimization process, employing&#xa0;Particle Swarm Optimization (PSO)&#xa0;and&#xa0;Bayesian Optimization (BO), was used to fine-tune hyperparameters. The findings suggest that the AdaBoost regressor outperforms other individual models in LI prediction, while RF is the most proficient model for predicting Su. Ensemble techniques, particularly WA, significantly enhanced model performance, achieving remarkable&#xa0;coefficient of determination (R<sup>2</sup>)&#xa0;values in S<sub>u</sub> (0.9913 for BO, 0.9961 for PSO) and LI prediction (0.9624 for BO, 0.9338 for PSO) during testing. Combining base learners and stacking with&#xa0;Support Vector Regression (SVR)&#xa0;further improved performance, surpassing other models. In the optimization comparison, BO demonstrated superior LI prediction, while PSO excelled in S<sub>u</sub> prediction, highlighting the distinctive strengths of each technique.&#xa0;SHAP analysis&#xa0;further identified the&#xa0;depth of cone penetration (d)&#xa0;as the most influential factor for both LI and S<sub>u</sub>, thereby enhancing model interpretability and demonstrating strong alignment with established geotechnical principles.</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

Integrating Bayesian and Particle Swarm Optimization in Ensemble Machine Learning for Enhanced Prediction of Soil Liquidity Index and Shear Strength

  • Mohammadreza Mahmoudi,
  • Giovanni Spagnoli,
  • Satoru Shimobe

摘要

This study presents a comprehensive exploration of the development of diverse stacking machine learning models for predicting the Soil Liquid Index (LI) and Soil Undrained Shear Strength (Su). The study utilizes a wide range of datasets, including over 1550 data points for LI and 950 for Su, collected from various research studies. Individual models such as Multilayer Perceptron (MLP), Random Forest (RF), and AdaBoost regressors were evaluated alongside ensemble models, including Simple Averaging (SA), Weighted Averaging (WA), and stacking-SVR. The optimization process, employing Particle Swarm Optimization (PSO) and Bayesian Optimization (BO), was used to fine-tune hyperparameters. The findings suggest that the AdaBoost regressor outperforms other individual models in LI prediction, while RF is the most proficient model for predicting Su. Ensemble techniques, particularly WA, significantly enhanced model performance, achieving remarkable coefficient of determination (R2) values in Su (0.9913 for BO, 0.9961 for PSO) and LI prediction (0.9624 for BO, 0.9338 for PSO) during testing. Combining base learners and stacking with Support Vector Regression (SVR) further improved performance, surpassing other models. In the optimization comparison, BO demonstrated superior LI prediction, while PSO excelled in Su prediction, highlighting the distinctive strengths of each technique. SHAP analysis further identified the depth of cone penetration (d) as the most influential factor for both LI and Su, thereby enhancing model interpretability and demonstrating strong alignment with established geotechnical principles.