This study uses as reference the conventional method of slope failure risk evaluation using a support vector machine, and assumes cases in which only a small number of the study data required for analysis is available. An attempt was made to propose a method of analysis using virtual data as a way of complementing such scarcity of training data.
As a result of analysis, it became possible to construct separated surfaces with high precision by avoiding the expansion of contradictory data in the preparation of the virtual data and setting the parameters of the support vector machine in such a way that the number of times of additions/expansions of the virtual data is made as many as possible. Further, by using separated surfaces obtained from these virtual data, it became possible to evaluate slopes for which previously judgments of safety/risk could not be made. In addition, it was confirmed by site surveys of these slopes that there is agreement between the findings obtained from actual site conditions and the results of analysis obtained in this study, indicating the usefulness of the method.
View full abstract