Assessing the Performance of Machine Learning Algorithms for Soil Classification Using Cone Penetration Test Data

Chala, Ayele Tesema; Ray, Richard P.

doi:10.3390/app13095758

Cited by 10 publications

(3 citation statements)

References 49 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…These modern tools offer more precise solutions and additional insights into soil behavior. Recent studies have provided valuable insights into the application of machine learning in geotechnical engineering [26][27][28][29][30]. These contributions contribute to a deeper awareness of incorporating machine learning techniques in geotechnical engineering.…”

Section: Introductionmentioning

confidence: 99%

Developing Nomographs for the Unit Weight of Soils

Dungca,

Galupino

2023

Buildings

View full text Add to dashboard Cite

Engineers have created increasingly complex correlations based on laboratory and field tests. Over time, geotechnical engineering modeling techniques have evolved from simple analytical methods to complex numerical modeling techniques. Nomographs are traditional computational tools that have been widely employed in engineering. Combining nomographs with computational tools such as numerical models and machine learning algorithms can lead to better outcomes. Thus, this study aimed to develop a nomograph for geotechnical engineering that incorporates machine learning, specifically for the unit weight of soil. Four calibrated models were developed to determine the unit weight of soil: the moist unit weight of coarse-grained soil, the saturated unit weight of coarse-grained soil, the moist unit weight of fine-grained soil, and the saturated unit weight of fine-grained soil. An uncertainty test was conducted for the data used. Our results indicated a strong positive relationship to most of the models. The generated nomographs were tested in Malabon, a city in Metro Manila, where a low unit weight of soil was determined. This low unit weight was validated by the predominance of alluvial deposits and the shallow groundwater table, which soften and weaken the soil.

show abstract

Section: Introductionmentioning

confidence: 99%

Developing Nomographs for the Unit Weight of Soils

Dungca,

Galupino

2023

Buildings

View full text Add to dashboard Cite

show abstract

“…This shift towards AI and ML re ects a desire for more e cient and reliable methods in geotechnical research and engineering practices. RF and Support Vector Machine (SVM) was used to classify soil from CPT test by ( Chala & Ray, 2023). The study used depth and effective stress as input features which were used to classify the soil but the soil unit weight was not used in this study.…”

Section: Introdutionmentioning

confidence: 99%

Use of CPT and other Parameters for Estimating Soil Unit Weight Using Optimised Machine Learning Models

Roy,

Abir,

Ansary

2024

Preprint

View full text Add to dashboard Cite

In this research, we explore the prediction of soil unit weight using five advanced machine learning algorithms: AdaBoost with Random Forest, Multi-Layer Perceptron (MLP), Random Forest (RF), Support Vector Regression (SVR) and XGBoost. Random Forest serves as a weak learner within AdaBoost. We optimize the hyperparameters of these algorithms using randomized search cross-validation (RSCV) and evaluate their performance using mean average error (MAE), root mean square error (RMSE), and R2 metrics. The input features consist of soil sample depth (D), moisture content (MC), fine content (FC), cone tip resistance (QC), and cone local resistance (FS). Additionally, we employ an autoencoder-based feature augmentation technique to enhance the models' ability to capture complex patterns in the data. Before feature augmentation, AdaBoost with Random Forest achieves the highest performance (R2 = 0.896), while SVR exhibits the lowest accuracy (R2 = 0.7402) on the test dataset. Post-augmentation, both AdaBoost with RF and SVR show improvements in R2, MAE and RMSE values, indicating that augmented features capture more variability. XGBoost, Random Forest, and Multi-Layer Perceptron rank 2nd, 3rd, and 4th, respectively, in terms of R2 value. SHAP analysis reveals that QC and FS negatively impact model accuracy, while FC and MC have both positive and negative effects. D emerges as the most influential feature contributing positively to model accuracy. In conclusion, AdaBoost with Random Forest yields the highest accuracy in predicting soil unit weight, with D being the most critical feature.

show abstract

“…Many ML algorithms, such as gradient boosting, random forest, support vector machine (SVM) artificial neural network (ANN), and decision trees (DT), have been used in various geotechnical applications, including soil classification [27][28][29][30][31][32][33], V s prediction [23][24][25][26]34], liquefaction analysis [35][36][37][38][39][40], stability analysis [41][42][43][44][45], and settlement prediction [46][47][48]. The application of ML algorithms in geotechnical engineering has shown promising results in terms of efficiency and accuracy.…”

Section: Introductionmentioning

confidence: 99%

Machine Learning Techniques for Soil Characterization Using Cone Penetration Test Data

Chala

Ray

2023

Applied Sciences

Self Cite

View full text Add to dashboard Cite

Seismic response assessment requires reliable information about subsurface conditions, including soil shear wave velocity (Vs). To properly assess seismic response, engineers need accurate information about Vs, an essential parameter for evaluating the propagation of seismic waves. However, measuring Vs is generally challenging due to the complex and time-consuming nature of field and laboratory tests. This study aims to predict Vs using machine learning (ML) algorithms from cone penetration test (CPT) data. The study utilized four ML algorithms, namely Random Forests (RFs), Support Vector Machine (SVM), Decision Trees (DT), and eXtreme Gradient Boosting (XGBoost), to predict Vs. These ML models were trained on 70% of the datasets, while their efficiency and generalization ability were assessed on the remaining 30%. The hyperparameters for each ML model were fine-tuned through Bayesian optimization with k-fold cross-validation techniques. The performance of each ML model was evaluated using eight different metrics, including root mean squared error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), coefficient of determination (R2), performance index (PI), scatter index (SI), A10−I, and U95. The results demonstrated that the RF model consistently performed well across all metrics. It achieved high accuracy and the lowest level of errors, indicating superior accuracy and precision in predicting Vs. The SVM and XGBoost models also exhibited strong performance, with slightly higher error metrics compared with the RF model. However, the DT model performed poorly, with higher error rates and uncertainty in predicting Vs. Based on these results, we can conclude that the RF model is highly effective at accurately predicting Vs using CPT data with minimal input features.

show abstract

Assessing the Performance of Machine Learning Algorithms for Soil Classification Using Cone Penetration Test Data

Cited by 10 publications

References 49 publications

Developing Nomographs for the Unit Weight of Soils

Developing Nomographs for the Unit Weight of Soils

Use of CPT and other Parameters for Estimating Soil Unit Weight Using Optimised Machine Learning Models

Machine Learning Techniques for Soil Characterization Using Cone Penetration Test Data

Contact Info

Product

Resources

About