Divesh Ranjan Kumar
Prediction of probability of liquefaction using soft computing techniques - Vol.103(4), Dec - USA Springer 2022 - 1195-1208p.
Prediction of liquefaction potential of any soil deposit is itself a very challenging task. The problem becomes even more demanding when it becomes necessary to incorporate the variability of all related parameters. Because the parameters that impact liquefaction potential are inherently unknown, the problem is probabilistic rather than deterministic. In the literature, probabilistic analysis of liquefaction potential has attracted a lot of attention, and it's been shown to be a useful technique for evaluating uncertainty inherent in the problem. Machine Learning (ML) techniques have found their applications in all fields of science and engineering while dealing with problems of stochastic nature. These techniques are capable of finding out the desired outputs very effectively. In this paper, five different ML models namely, extreme gradient boosting (XGBoost), random forest (RF), gradient boosting machines (GBM), support vector regression (SVR), and group method of data handling (GMDH) have been used for evaluation of probability of liquefaction based on standard penetration test data. In this study, analysis has been carried out with six input variable such as, depth of penetration, corrected standard penetration blow number, total vertical stress, fine content, maximum horizontal acceleration, total effective stress, and earthquake magnitude. To examine the capabilities of the suggested models in predicting the probability of liquefaction, several statistical parameters have been examined. To compare the accuracy of the proposed models, Taylor graph, REC curve, and error matrix have been developed. While all of the proposed models could efficiently predict the probability of liquefaction. XGBoost model has been found to give the best prediction among all five models. In summary, XGBoost model attained (R2=0.978 for training and R2=0.799 for testing), GBM model attained (R2=.953 for training and R2=0.780 for testing), RF model attained (R2=.930 for training and R2=0.769 for testing), SVR model attained (R2=.702. for training and R2=0.778 for testing), GMDH model attained (R2=0.650 for training and R2=0.701 for testing). The proposed models can also be utilized as a valid model for forecasting the probability of liquefaction efficiently for complicated real-world earthquake engineering problems.
Humanities and Applied Sciences
Prediction of probability of liquefaction using soft computing techniques - Vol.103(4), Dec - USA Springer 2022 - 1195-1208p.
Prediction of liquefaction potential of any soil deposit is itself a very challenging task. The problem becomes even more demanding when it becomes necessary to incorporate the variability of all related parameters. Because the parameters that impact liquefaction potential are inherently unknown, the problem is probabilistic rather than deterministic. In the literature, probabilistic analysis of liquefaction potential has attracted a lot of attention, and it's been shown to be a useful technique for evaluating uncertainty inherent in the problem. Machine Learning (ML) techniques have found their applications in all fields of science and engineering while dealing with problems of stochastic nature. These techniques are capable of finding out the desired outputs very effectively. In this paper, five different ML models namely, extreme gradient boosting (XGBoost), random forest (RF), gradient boosting machines (GBM), support vector regression (SVR), and group method of data handling (GMDH) have been used for evaluation of probability of liquefaction based on standard penetration test data. In this study, analysis has been carried out with six input variable such as, depth of penetration, corrected standard penetration blow number, total vertical stress, fine content, maximum horizontal acceleration, total effective stress, and earthquake magnitude. To examine the capabilities of the suggested models in predicting the probability of liquefaction, several statistical parameters have been examined. To compare the accuracy of the proposed models, Taylor graph, REC curve, and error matrix have been developed. While all of the proposed models could efficiently predict the probability of liquefaction. XGBoost model has been found to give the best prediction among all five models. In summary, XGBoost model attained (R2=0.978 for training and R2=0.799 for testing), GBM model attained (R2=.953 for training and R2=0.780 for testing), RF model attained (R2=.930 for training and R2=0.769 for testing), SVR model attained (R2=.702. for training and R2=0.778 for testing), GMDH model attained (R2=0.650 for training and R2=0.701 for testing). The proposed models can also be utilized as a valid model for forecasting the probability of liquefaction efficiently for complicated real-world earthquake engineering problems.
Humanities and Applied Sciences