Search

Soil liquefaction is a primary factor in causing destruction during earthquakes. For over four decades, there has been significant improvement in detecting soil liquefaction. At initially, this process was primarily in sandy, clean soils. Because the cost of soil transformation is often significant, exact estimate of liquefaction potential, together with security considerations, might reduce the scheme's fiscal cost. This research proposes to provide a novel soil liquefaction prediction model with three primary stages. It includes, data visualization was performed using correlation matrix and pair plots to determine the dependency and independency of each variable, as well as the entropy of the data to determine the complexity of the data, before deploying a novel liquefaction methodology that included an ensemble model of sophisticated deep learning classifiers of Long short-term memory (LSTM) + Support Vector Machines (SVM) to reduce the reproducibility problem. Improved Correlation characteristics have been used to pick the most essential variables while removing duplicate and unnecessary characteristics. K-fold validation is used to prevent overfitting, a situation when a model is excessively trained on training data but underperforms on new, untested data. Finally, the grey wolf optimizer was used to improve the operation's local minimum values and convergence. • Proposed a novel soil liquefaction prediction model that integrates deep learning Long short-term memory (LSTM) + Support Vector Machines (SVM) and Gray Wolf Optimizer. • An improved correlation strategy is proposed to identify key variables. • K-fold validation is used, which ensures robustness by mitigating overfitting on unseen data. • Incorporation of a novel approach achieved better performances, especially in false negative rates (FNR). [ABSTRACT FROM AUTHOR]