Liquefaction Potential Assessment of Soils Using Machine Learning Techniques: A State-of-the-Art Review from 1994–2021

Machine learning (ML) has emerged as a powerful tool for prediction of systems behavior in many engineering disciplines. A few applications of ML techniques are available in geotechnical engineering and other fields of civil engineering. The existing review studies on the application of ML techniques in conventional geotechnical engineering and earthquake geotechnics are available in the broader areas but not specific to liquefaction phenomenon. Studies exist on the liquefaction potential of cohesionless soils using ML techniques with a varying degree of success. More studies are needed to formalize the use of ML techniques in the seismic liquefaction assessment. In this review, an attempt is made to critically review the existing literature on ML techniques as applied to the liquefaction analysis. The published literature from 1994 to 2021 has been collected, critically reviewed, and presented systematically in the form of easy to understand tables and figures. The tables are labeled based on the data requirement in ML techniques, conventional methods, and in-situ tests. A summary table highlights the relative importance of input variables in the dataset required for the liquefaction potential assessment. Limitations of conventional methods and existing ML models, the importance of a large database, relative comparison of the four developed ML models, and the requirement of probabilistic analysis are included. A gap in the existing literature is outlined and followed by a way forward for future research. It is concluded that there is a need to update the liquefaction database and modify existing ML algorithms so that they become computationally efficient and reliable for liquefaction potential assessment.