Application of Artificial Intelligence to Cluster Soil Behaviour from CPTu Data

In this study, the authors used many artificial intelligence algorithms to cluster soil behaviour from CPTu data, that includes cone resistance (qc), frictional resistance (fs), dynamic pore pressure (u2), corrected cone resistance (qt) and friction ratio (Rf). The soil behavior type is following Robertson (1986). There are four model are built in this study include: i. Supervised learning with SVM algorithm by qc, fs, u2, qt and Rf; ii. Supervised learning with SVM algorithm by qt and Rf; iii. Unsupervised learning with Kmeans algorithm by qt and Rf with three clusters; and Unsupervised learning with Kmeans algorithm by qt and Rf with nine clusters. To satisfy “Imbalanced data” and Roberson’s chart shape, the raw data are being preprocessing within two steps before clustering with KMeans or continue to divide to 3 minor set for SVM algorithm that includes: training set - 50%, validation set - 20% and test set - rest. The result indicated the Supervised learning with SVM algorithm by qc, fs, u2, qt and Rf is the best model, while unsupervised learning with KMeans does not meet the requirements.