2026 Volume 7 Issue 3 Pages 326-338
To properly design the layout of horizontal drainage boreholes in road embankments, it is necessary to identify the factors that contribute to drainage effectiveness and to construct a model to predict discharge volume. To this end, focusing on five embankments in the Tohoku region, we conducted a multiple regression analysis on 88 boreholes to extract explanatory variables for the discharge volume from the drainage boreholes. Using the obtained variables, a logistic regression analysis was then performed. Furthermore, TabPFN (Tabular Foundation Model), which is capable of handling sparse tabular data, was used to evaluate its performance in classifying the presence or absence of borehole discharge. While the logistic regression demonstrated high classification performance with an ROC-AUC of 0.963, TabPFN yielded an ROC-AUC of 0.924, a precision of 0.846, and an F1 score of 0.830. A remaining challenge of the current model is that it does not account for factors such as the time lag in groundwater base flow in response to rainfall history.