Evaluation of liquefaction potential in central Taiwan using random forest method

Area of Science:

• Geotechnical Engineering
• Seismology
• Machine Learning Applications

Background:

• Soil liquefaction poses a significant geotechnical hazard in seismically active regions, impacting infrastructure and public safety.
• Accurate prediction of liquefaction potential is crucial for seismic risk assessment and mitigation strategies.

Purpose of the Study:

• To evaluate the liquefaction potential in central Taiwan using the random forest (RF) machine learning method.
• To compare the performance of the RF model against conventional simplified procedures for liquefaction prediction.

Main Methods:

• Development of RF models using a dataset of 540 soil and seismic parameters.
• Inclusion of factors such as depth, stresses, SPT-N values, fine content, earthquake magnitude, and peak ground acceleration.
• Rigorous validation using cross-validation and comparison with historical liquefaction data.

Main Results:

• The RF model achieved a high prediction accuracy of 98.89%.
• SPT-N value identified as the most critical soil factor, and peak ground acceleration as the key seismic factor.
• The RF model demonstrated superior performance compared to simplified procedures, even with fewer input variables.

Conclusions:

• The random forest method is highly effective for predicting soil liquefaction potential.
• The developed RF model offers a more accurate and robust approach to seismic hazard assessment than conventional methods.
• This research provides a valuable tool for geotechnical engineers and urban planners in seismically prone areas.