Tailored clustering-based bayesian model for GEOAI benchmark problems BM/AirportSoilProperties/2/2025
Year: 2026
Type: article
Abstract: BM/AirportSoilProperties/2/2025 presents two benchmark problems with a verification site located in a large-scale offshore airport in Japan involving the prediction of: (1) spatial variation of undrained shear strength; (2) missing mechanical parameters. For both benchmark problems, Big Indirect Data (BID) can be used to reduce the bias and the uncertainty in the predictions. In this paper, BID is defined as a geo-database containing test records from many boreholes that are not directly applicable to the locations of interest. This paper adopts a tailored clustering-based Bayesian model to solve the two benchmark problems (the proposed solution). Results show that the 95% confident intervals of soil properties predicted by the proposed solution generally encompass the true unobserved data (the validation data), demonstrating the reasonableness of the predictions. The performance of the proposed solution is further evaluated through comparison of root mean square error (RMSE), modified root mean square error (RMSE2), and runtime with the sample solutions introduced in BM/AirportSoilProperties/2/2025. A lower RMSE or RMSE2 indicates lower prediction uncertainty and bias. In most cases, the proposed solution achieves the lowest RMSE and RMSE2 values with the shortest runtimes, outperforming the sample solutions.
Source: Geodata and AI.
Authors: Yongmin Cai, Kok-Kwang Phoon
Institution: Singapore University of Technology and Design
Language: English
Subfield: Safety, Risk, Reliability and Quality
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $2,000
Award: G4625344686
OpenAlex ID: W7130510797