Patient rule-induction method for liquefaction potential assessment based on CPT data

Determination of seismic liquefaction susceptibility of soils is an imperative task in earthquake science. In the present study, the patient ruled-induction method (PRIM), for the first time in the engineering literature, has been successfully employed to investigate the potential of liquefaction occurrence due to earthquake strikes in soils. To develop the PRIM model, a comprehensive database including 444 cone penetration tests field observations from different sources in the literature is collected. Six different predictive variables that characterize soil and earthquake parameters are considered as input variables to identify sub-regions where liquefaction phenomena may occur. The PRIM algorithm discovered very high-risk validated sub-regions with at least 92% morality, each covering at least 5% of the total instances. The performance of the developed subgroups is compared to that of the developed ANN, SVM, and ANFIS models. Results confirmed the robustness and superiority of the PRIM against these algorithms. The main advantages of the PRIM are its capability to produce simple and easy-to-interpret models and also to result in a probability space. These suggest its applicability to geotechnical and other engineering failure problems.