Solving BM/AirportSoilProperties/2/2025 using simple machine learning algorithms

This paper presents solutions for GEOAI benchmark problems BM/AirportSoilProperties/2/2025. Two simple machine learning (ML) methods, l 1 mixed filtering and Bayesian Linear Regression (BLR), were used to develop a new site characterization method. To demonstrate its effectiveness, the unknown undrained shear strength s u in the cluster-BID/4 was predicted. The revised root-mean-square error (RMSE) was also proposed to more reasonably evaluate the performance of ML predictions. Although the RMSE values obtained using the proposed method are larger than those reported in the sample solutions by Otake et al. (2025), the predictive probability distributions of undrained shear strength s u still reasonably capture the true s u values. This demonstrates that even simple ML methods can perform well in challenging benchmark problems for site characterization. The results also highlight the importance of adaptively selecting a suitable BID for the target dataset.