Advancements in predicting soil liquefaction susceptibility: a comprehensive analysis of ensemble and deep learning approaches
Contact us if these videos are not relevant.
Contact us if these videos are not relevant.
View abstract on PubMed
Summary
This summary is machine-generated.This study evaluated deep learning (DL) and ensemble learning models for soil liquefaction susceptibility assessment. The BI-LSTM model demonstrated the highest accuracy, offering robust predictions for seismic risk mitigation.
Area Of Science
- Geotechnical Engineering
- Earthquake Engineering
- Machine Learning Applications
Background
- Liquefaction, a loss of soil strength due to increased pore water pressure, poses significant seismic risks.
- Accurate assessment of liquefaction susceptibility is crucial for construction site safety and seismic hazard mitigation.
- Traditional methods may not fully capture the complex factors influencing liquefaction.
Purpose Of The Study
- To investigate the effectiveness of deep learning (DL) and ensemble learning models for assessing soil liquefaction susceptibility.
- To compare the performance of various machine learning models using a comprehensive database of cone penetration test (CPT) data and historical earthquake observations.
- To identify the most reliable model for predicting liquefaction potential.
Main Methods
- Utilized a large database of cone penetration test (CPT) measurements and field liquefaction performance observations.
- Developed and evaluated deep learning (DL) models, including BI-LSTM and LSTM, and ensemble learning models like XGBoost and Random Forest (RF).
- Assessed model performance using multiple metrics: accuracy, validation loss, precision, recall, F1 score, MCC, BA, ROC curve analysis, and AUC.
Main Results
- The BI-LSTM model achieved the highest accuracy (0.9791 training, 0.8889 testing), indicating strong predictive ability and generalizability.
- LSTM and XGBoost also demonstrated strong performance, with accuracies of 0.9433/0.8750 and 0.9194/0.8750, respectively.
- The Random Forest (RF) model showed a notable difference between training (0.9373) and testing (0.8681) accuracies.
Conclusions
- BI-LSTM is the most reliable model for assessing liquefaction potential, followed closely by LSTM and XGBoost.
- DL models (BI-LSTM, LSTM) excel at capturing sequential dependencies, while ensemble models (XGBoost, RF) offer robust results from structured data.
- This research contributes advanced tools for liquefaction hazard assessment, improving seismic risk mitigation strategies.
Contact us if these videos are not relevant.
Contact us if these videos are not relevant.
Related Concept Videos
The dynamic modulus of elasticity assesses how a concrete structure deforms under impact or dynamic loads. It is typically higher than the static modulus of elasticity, measured under slow, steady loading conditions.
The sonic test is a common method to determine the dynamic modulus. In this test, a concrete beam, sized either 6 x 6 x 30 inches or 4 x 4 x 20 inches, is clamped at its center. Vibrations are initiated at one end of the beam by an electromagnetic exciter unit powered by...
Upon subjecting concrete to moderate or high uniaxial compressive or tensile stresses, the strain response is non-linear relative to the stress applied. As the stress is removed, the resulting stress-strain curve deviates from the original path traced during loading, creating a hysteresis loop, indicative of the concrete's non-linear and non-elastic properties. Typically, a material's modulus of elasticity, which is a measure of the material's stiffness, is inferred from the linear...