TY - JOUR
T1 - Prediction of Liquefied Soil Settlement Using Multilayer Perceptron with Bayesian Optimization
AU - Van Nguyen, Nhi
AU - Van Le, Luong
AU - Nguyen, Tan No
AU - Park, Sung Sik
AU - Tran, Thanh Danh
N1 - Publisher Copyright:
© The Author(s), under exclusive licence to Indian Geotechnical Society 2024.
PY - 2024/12
Y1 - 2024/12
N2 - Liquefaction-induced settlement in saturated loose soils is a commonly geotechnical phenomenon resulted from the loss of its shear strength under moderate or high seismicity. This can have detrimental effects on building, infrastructure, and even human life. Therefore, predicting the settlement of ground caused by liquefaction plays a critical role in the geotechnical field. As the limited data for liquefaction-induced settlement obtained, an effective method should be applied to provide highly accurate prediction. The aim of this study is to propose a machine learning approach, namely multilayer perceptron (MLP), with a wide range of hyperparameters optimized by the Bayesian optimization method to predict liquefaction-induced settlement due to the Pohang earthquake in South Korea. The ground settlement and its correlation with different soil conditions were taken into account including unit weight, soil layer depth, standard penetration test blow count, and cyclic stress ratio. Moreover, other five well-known models, namely linear regression, support vector machine, robust regression, and polynomial regression, were performed to facilitate a comparison study. The experimental result indicates that while the high influence of all considerable variables on the ground settlement was obtained from the correlation coefficients, the soil layer depth factor had the highest one. Based on the Bayesian optimization, a greater prediction accuracy was observed from the proposed MLP model compared to other methods by using the evaluation metric of R-squared value. Furthermore, the developed model outperforms other machine learning methods proposed by previous studies in terms of predicting the soil settlement caused by liquefaction.
AB - Liquefaction-induced settlement in saturated loose soils is a commonly geotechnical phenomenon resulted from the loss of its shear strength under moderate or high seismicity. This can have detrimental effects on building, infrastructure, and even human life. Therefore, predicting the settlement of ground caused by liquefaction plays a critical role in the geotechnical field. As the limited data for liquefaction-induced settlement obtained, an effective method should be applied to provide highly accurate prediction. The aim of this study is to propose a machine learning approach, namely multilayer perceptron (MLP), with a wide range of hyperparameters optimized by the Bayesian optimization method to predict liquefaction-induced settlement due to the Pohang earthquake in South Korea. The ground settlement and its correlation with different soil conditions were taken into account including unit weight, soil layer depth, standard penetration test blow count, and cyclic stress ratio. Moreover, other five well-known models, namely linear regression, support vector machine, robust regression, and polynomial regression, were performed to facilitate a comparison study. The experimental result indicates that while the high influence of all considerable variables on the ground settlement was obtained from the correlation coefficients, the soil layer depth factor had the highest one. Based on the Bayesian optimization, a greater prediction accuracy was observed from the proposed MLP model compared to other methods by using the evaluation metric of R-squared value. Furthermore, the developed model outperforms other machine learning methods proposed by previous studies in terms of predicting the soil settlement caused by liquefaction.
KW - Bayesian optimization
KW - Liquefaction
KW - Machine learning
KW - Multilayer perceptron
KW - Settlement
UR - http://www.scopus.com/inward/record.url?scp=85185950191&partnerID=8YFLogxK
U2 - 10.1007/s40098-024-00894-w
DO - 10.1007/s40098-024-00894-w
M3 - Article
AN - SCOPUS:85185950191
SN - 0971-9555
VL - 54
SP - 2443
EP - 2453
JO - Indian Geotechnical Journal
JF - Indian Geotechnical Journal
IS - 6
ER -