Abstract

We present a novel method using four artificial intelligence (AI) algorithms to anticipate the cumulative degree of soil compaction (CDSC) after dynamic compaction (DC). Four AI algorithms adopted in this study include support vector regression SVR, artificial neural network (ANN), random forest (RF), and gradient boosting machine (GBM). Input variables for AI algorithms involve the average SPT N-value before dynamic compaction, cumulative applied energy normalized with a cross-sectional area of tamper, and the number of the tamper drops. Apart from cross-validation with a testing set, additional in situ test data compiled from a different section within the studied site are used to estimate the generalized capacity of the AI models. In addition, we conduct out-of-distribution analyses for the four AI algorithms in view of parametric studies. The CDSC prediction performance for the four AI models results in high prediction metrics of accuracy with the r2 higher than 0.9 for the testing scenario while the r2 of the other AI models is more than 0.9 when out-of-sample data are considered except for the GBM. The ANN seems to be the best model as the parametric study considers out-of-distribution data and suggests a strong relationship between input variables and CDSC that is more coherent with engineering principles for DC. Finally, the ANN model can be utilized to develop a mathematical model for CDSC prediction.

Abstract

We present a novel method using four artificial intelligence (AI) algorithms to anticipate the cumulative degree of soil compaction (CDSC) after dynamic compaction (DC). Four AI algorithms adopted in this study include support vector regression SVR, artificial neural network (ANN), random forest (RF), and gradient boosting machine (GBM). Input variables for AI algorithms involve the average SPT N-value before dynamic compaction, cumulative applied energy normalized with a cross-sectional area of tamper, and the number of the tamper drops. Apart from cross-validation with a testing set, additional in situ test data compiled from a different section within the studied site are used to estimate the generalized capacity of the AI models. In addition, we conduct out-of-distribution analyses for the four AI algorithms in view of parametric studies. The CDSC prediction performance for the four AI models results in high prediction metrics of accuracy with the r2 higher than 0.9 for the testing scenario while the r2 of the other AI models is more than 0.9 when out-of-sample data are considered except for the GBM. The ANN seems to be the best model as the parametric study considers out-of-distribution data and suggests a strong relationship between input variables and CDSC that is more coherent with engineering principles for DC. Finally, the ANN model can be utilized to develop a mathematical model for CDSC prediction.