Predicting seismic-induced liquefaction through ensemble learning frameworks.

Area of Science:

• Geotechnical Engineering
• Machine Learning
• Seismology

Background:

• Liquefaction prediction is challenging due to regional data limitations and reliance on indirect methods.
• Machine learning offers direct prediction capabilities, with ensemble learning representing a recent advancement.
• Existing literature lacks comprehensive evaluations of ensemble learning for liquefaction prediction.

Purpose of the Study:

• To investigate the effectiveness of ensemble learning approaches for predicting seismic-induced liquefaction.
• To address the need for comprehensive evaluation of ensemble learner generalization abilities in this domain.
• To explore advanced machine learning techniques for improved geotechnical hazard assessment.

Main Methods:

• Conducted a comprehensive analysis of fifteen distinct ensemble learning models.
• Applied ensemble learning frameworks to the binary classification problem of liquefaction occurrence.
• Trained multiple learners within predefined ensemble architectures and integrated their inferences.

Main Results:

• Ensemble models demonstrated superior prediction performance over single machine learning models.
• The application of ensemble learning led to a significant reduction in prediction uncertainty.
• Fifteen ensemble models were evaluated, highlighting their potential in geotechnical applications.

Conclusions:

• Ensemble learning provides a powerful and effective direct prediction method for seismic liquefaction.
• The study confirms the benefits of ensemble methods in enhancing generalization ability and reducing uncertainty.
• Further research into ensemble learning is warranted for advancing liquefaction hazard analysis.