Classification Artificial Intelligence Algorithms Coupled with Clustering Algorithms: A Potential Technique to Consider for Predicting Soil Liquefaction Phenomena

Predictions of soil liquefaction phenomena under seismic loads have been investigated through available deterministic and statistical methods. However, it isn’t easy to develop a model considering all the independent soil variables. To overcome this limitation, this paper demonstrates that combining two artificial intelligence (AI) techniques could be worthwhile to assess soil liquefaction susceptibility based on the cone penetration test (CPT) data. So far, there has never been a study that used AI algorithms for soil clustering in liquefaction problems. Only a few studies compared the usability and performance of the prediction algorithms. At first, the data is clustered, and then only the cluster representatives are used for the prediction algorithms—the first technique uses the k-medoids clustering algorithm. The second uses AI algorithms such as artificial neural network (ANN), support vector machine (SVM), and random forest decision trees (RFDT) to predict the liquefaction susceptibility of soil. All three classification models indicate that the liquefaction susceptibility of the incompressible deposits with depthless critical layer is high compared to other clusters. This paper has demonstrated the usefulness of the clustering approach. The clustering model presented in this study can be a shortcut and simpler to apply than the conventional AI methods based on classification algorithms. The findings of the present work contribute to the artificial intelligence scientific revolution in the scope of the prediction models.