Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Implementing ensemble learning models for the prediction of shear strength of soil
https://orcid.org/http://orcid.org/0000-0001-8536-5305
https://orcid.org/http://orcid.org/0000-0003-2906-6479
https://orcid.org/http://orcid.org/0000-0002-4446-673X
Precise estimation of the shear strength (SS) parameters of soil is one of the important criteria in the geotechnical design process. The aim of this research is to find out the performance of five ensemble models, namely: (i) random forest (RF), (ii) extra trees regressor (ETR), (iii) light gradient boosting machine (LGBM), (iv) Adaboost regressor (ABR), and (v) gradient boosting regressor (GBR), for the prediction capacity of shear strength of soil (SSS). Based on the soil dataset consisting of 249 soil samples and 12 influencing factors, a common approach is formed to predict SSS with the help of the models considered in the research. The datasets for the testing and training phases of the model's generation and evaluation were created using the acquired data. The verification and assessment of models were executed with the help of the coefficient of determination (R2), root mean square log error (RMSLE), mean absolute error (MAE), mean squared errors (MSE), mean absolute percentage error (MAPE), and root mean square errors (RMSE) statistics. In order to determine the significance of various influencing parameters in the outcome prediction, a sensitivity analysis was also carried out. After the development and control of all five models, the best models were selected. Validation results showed that the ETR model has achieved a good modeling outcome with (MAE = 0.0289), (MSE = 0.0015), (RMSE = 0.0378), (R2 = 0.7934), (RMSLE = 0.0285), and (MAPE = 0.0900) outperforming other models. Hence, the model ETR produced better results with high accuracy and low error for the prediction of SSS. Also, the learning process of the best model was performed with the help of a model-based residual plot, a prediction error plot, a feature importance plot, Cook’s distance outlier detection plot, and a learning curve. The most powerful features that contributed to the prediction of SSS in the best model are the liquidity index (LI), water content (ω), void ratio (e), bulk density (ρb), and sample depth.
Implementing ensemble learning models for the prediction of shear strength of soil
https://orcid.org/http://orcid.org/0000-0001-8536-5305
https://orcid.org/http://orcid.org/0000-0003-2906-6479
https://orcid.org/http://orcid.org/0000-0002-4446-673X
Precise estimation of the shear strength (SS) parameters of soil is one of the important criteria in the geotechnical design process. The aim of this research is to find out the performance of five ensemble models, namely: (i) random forest (RF), (ii) extra trees regressor (ETR), (iii) light gradient boosting machine (LGBM), (iv) Adaboost regressor (ABR), and (v) gradient boosting regressor (GBR), for the prediction capacity of shear strength of soil (SSS). Based on the soil dataset consisting of 249 soil samples and 12 influencing factors, a common approach is formed to predict SSS with the help of the models considered in the research. The datasets for the testing and training phases of the model's generation and evaluation were created using the acquired data. The verification and assessment of models were executed with the help of the coefficient of determination (R2), root mean square log error (RMSLE), mean absolute error (MAE), mean squared errors (MSE), mean absolute percentage error (MAPE), and root mean square errors (RMSE) statistics. In order to determine the significance of various influencing parameters in the outcome prediction, a sensitivity analysis was also carried out. After the development and control of all five models, the best models were selected. Validation results showed that the ETR model has achieved a good modeling outcome with (MAE = 0.0289), (MSE = 0.0015), (RMSE = 0.0378), (R2 = 0.7934), (RMSLE = 0.0285), and (MAPE = 0.0900) outperforming other models. Hence, the model ETR produced better results with high accuracy and low error for the prediction of SSS. Also, the learning process of the best model was performed with the help of a model-based residual plot, a prediction error plot, a feature importance plot, Cook’s distance outlier detection plot, and a learning curve. The most powerful features that contributed to the prediction of SSS in the best model are the liquidity index (LI), water content (ω), void ratio (e), bulk density (ρb), and sample depth.
Implementing ensemble learning models for the prediction of shear strength of soil
https://orcid.org/http://orcid.org/0000-0001-8536-5305
https://orcid.org/http://orcid.org/0000-0003-2906-6479
https://orcid.org/http://orcid.org/0000-0002-4446-673X
Precise estimation of the shear strength (SS) parameters of soil is one of the important criteria in the geotechnical design process. The aim of this research is to find out the performance of five ensemble models, namely: (i) random forest (RF), (ii) extra trees regressor (ETR), (iii) light gradient boosting machine (LGBM), (iv) Adaboost regressor (ABR), and (v) gradient boosting regressor (GBR), for the prediction capacity of shear strength of soil (SSS). Based on the soil dataset consisting of 249 soil samples and 12 influencing factors, a common approach is formed to predict SSS with the help of the models considered in the research. The datasets for the testing and training phases of the model's generation and evaluation were created using the acquired data. The verification and assessment of models were executed with the help of the coefficient of determination (R2), root mean square log error (RMSLE), mean absolute error (MAE), mean squared errors (MSE), mean absolute percentage error (MAPE), and root mean square errors (RMSE) statistics. In order to determine the significance of various influencing parameters in the outcome prediction, a sensitivity analysis was also carried out. After the development and control of all five models, the best models were selected. Validation results showed that the ETR model has achieved a good modeling outcome with (MAE = 0.0289), (MSE = 0.0015), (RMSE = 0.0378), (R2 = 0.7934), (RMSLE = 0.0285), and (MAPE = 0.0900) outperforming other models. Hence, the model ETR produced better results with high accuracy and low error for the prediction of SSS. Also, the learning process of the best model was performed with the help of a model-based residual plot, a prediction error plot, a feature importance plot, Cook’s distance outlier detection plot, and a learning curve. The most powerful features that contributed to the prediction of SSS in the best model are the liquidity index (LI), water content (ω), void ratio (e), bulk density (ρb), and sample depth.
Implementing ensemble learning models for the prediction of shear strength of soil
Asian J Civ Eng
Rabbani, Ahsan (Autor:in) / Samui, Pijush (Autor:in) / Kumari, Sunita (Autor:in)
Asian Journal of Civil Engineering ; 24 ; 2103-2119
01.11.2023
17 pages
Aufsatz (Zeitschrift)
Elektronische Ressource
Englisch
Utilization of Tree-Based Ensemble Models for Predicting the Shear Strength of Soil
| Springer Verlag | 2024
Utilization of Tree-Based Ensemble Models for Predicting the Shear Strength of Soil
| Springer Verlag | 2024