Comparative and parametric study of AI-based models for risk assessment against soil liquefaction for high-intensity earthquakes.

An effort has been made to perform a detailed characterization of fine-grained soils considering the relative significance of plasticity index for liquefaction risk assessment of the alluvial soil along Indo-Gangetic River banks. To establish this, the author emphasizes the application of computational models in comparison to statistical or empirical approaches for investigation. The novel approach for evaluating the liquefaction risk of fine-grained deposits is performed using different computational models which were satisfactorily validated in the present study. These models consider SPT N-value and peak ground acceleration due to earthquakes along with the consideration of basic geotechnical properties of fine-grained soil. The comparative study suggests that the artificial neural network (ANN) model has achieved the best predictive accuracy as compared to the other four models and it can be used to facilitate the liquefaction risk assessment of fine-grained soil deposits. In addition, liquefaction hazard mapping for the concerned study area is presented based on the ANN model. The artificial intelligence (AI)-based liquefaction mapping technique has high potential as it radically advances risk prediction and preparedness against such seismic events. Also, the influence of some of the significant input parameters on liquefaction susceptibility has been elaborately discussed, in terms of the correlation matrix and relative significance plots. Liquid limit and plasticity index have achieved a relative weightage of 14% and 16%, respectively, for predicting factor of safety (F_L), whereas the SPT N-value has the highest impact of 21%. It is also observed that when the magnitude of the earthquake (M_w) is increased substantially from 6 to 7, a severe reduction in F_L of 27.04% was noted, whereas for increment in M_w from 7 to 8, 30.5% average reduction in F_L was observed. The finding of this study makes a substantial contribution in the field of liquefaction studies using AI models for fine-grained soil with medium to high plasticity. [ABSTRACT FROM AUTHOR]