Soil Categorization and Liquefaction Prediction Using Deep Learning and Ensemble Learning Algorithms

The study explores the use of machine learning techniques for accurate and reliable liquefaction predictions across four classes of soil types SM-SP, MI-ML-SC, CL-MH, and CH-CI. Deep learning and ensemble learning algorithms, namely Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), Extreme Gradient Boosting (XGB), and CatBoost (CatB), are incorporated to achieve this. The results of the study find that LSTM is the most accurate model, outperforming CNN, XGB, and CatB with an amazing accuracy of 0.96, precision of 0.95, recall of 0.94, and an F1-score of 0.95. These results demonstrate the superior capability of the LSTM model in capturing temporal dependencies and making accurate predictions. The finding of the study also underscores the level of liquefaction threat is highest in SM-SP soil, followed by MI-ML-SC, CL-MH, and CH-CI. By analyzing the liquefaction characteristics of multiple soil types, the study provides geotechnical engineers with an extensive framework for liquefaction evaluation. The incorporation of these sophisticated techniques provides a significant advancement in the accurate and reliable prediction of soil liquefaction potential, offering practical implications for geotechnical engineering.