JIAN-YE CHINGLi, XiangXiangLiFarahbakhsh, Hassan KamyabHassan KamyabFarahbakhsh2025-11-202025-11-20202500083674https://www.scopus.com/record/display.uri?eid=2-s2.0-105016843704&origin=resultslisthttps://scholars.lib.ntu.edu.tw/handle/123456789/733864The current study compiles a database named CPT-USCS/3/2017 that consists of 2017 pairwise cone penetration test (CPT) versus Unified Soil Classification System (USCS) category data from 228 global sites. The current study also proposes a novel hierarchical Bayesian model (HBM) framework named USCS-HBM to learn the inter-site and intra-site characteristics in the database. The USCS-HBM trained by the database can produce a prior model for the target site, and this prior model is updated by the sparse target-site data into the quasi-site-specific model. The resulting quasi-site-specific model can be adopted to predict USCS categories based on CPT measurements. The proposed USCS-HBM framework explicitly addresses the challenge of site uniqueness in CPT-based soil classification as well as the practical challenge of sparse target-site data. Case studies and extensive cross-validations showed that the proposed USCS-HBM framework can provide meaningful prediction results for USCS categories based on CPT measurements even if the target-site data are sparse.falseCPTdata-centric geotechnicshierarchical Bayesian modelsite uniquenesssoil classification[SDGs]SDG15journal article10.1139/cgj-2025-04132-s2.0-105016843704