Prediction of interfaces of geological formations using the multivariate adaptive regression spline method

The design and construction of underground structures are significantly affected by the distribution geological formations. Prediction interfaces using limited data has been a difficult task. A multivariate adaptive regression spline (MARS) method capable modeling nonlinearities automatically was used in this study to spatially predict elevations interfaces. Borehole from two sites Singapore were evaluate capability MARS for predicting By comparing predicted values with borehole data, it is shown that mean root square error 4.4 m Kallang Formation–Old Alluvium interface. In addition, able produce reasonable prediction intervals sense percentage testing covered 95% close associated confidence level, 95%. More importantly, interval evaluated had non-constant width appropriately reflected density complexity.