Improving Soil Liquefaction Potential Evaluation through AI-Based Prediction of SPT Values: A Comparative Study

1. Home
2. Publications
3. Online Library
4. Improving Soil Liquefaction Potential Evaluation through AI-Based Prediction of SPT Values: A Comparative Study

Liquefaction of soil is one of the most dangerous phenomena that can significantly impact civil structures and cause financial and human losses. The Standard Penetration Test (SPT) is one of the methods for evaluating soil liquefaction potential. However, despite the importance of this test, its uncertainty and the diversity of factors influencing it have made it difficult to predict. To address these challenges, this study attempts to predict the SPT value based on mechanical tests and artificial intelligence (AI) methods, namely support vector regression (SVR), and classification and regression random forests (CRRF). Models were developed using four inputs, including liquid limit, plasticity index, cohesion and friction angle, and the SPT as an output parameter. While it is true that SPT correction factors play a role in determining the final SPT value, the study specifically aimed to investigate the relationship between geotechnical parameters and SPT values. A statistical method, namely multiple linear regression (MLR). The researchers sought to gain insights into the complex nature of soil liquefaction and assess the effectiveness of AI techniques in predicting the SPT values, which are indicators of liquefaction potential. The comparison of the AI models allowed for an evaluation of their respective prediction performances and an understanding of how different techniques could potentially improve accuracy in assessing liquefaction potential. According to the results, the MLR model predicted the SPT value with coefficient of determination (R²) of 0.254 and the mean absolute error (MAE) of 3.353, whereas the CRRF and SVR models predicted SPT with R² 0.922 and 0.978, and MAE of 0.971 and 0.476. Moreover, the sensitivity analysis conducted on the best CRRF model suggested that the friction angle parameter has the greatest impact on all three models.

10th European Conference on Numerical Methods in Geotechnical Engineering (NUMGE2023)

6. Machine learning and artificial intelligence