Feature contributions in a high-risk area model for constructing a database to estimate landslide susceptibility

Various input parameters are required to estimate landslide susceptibility. However, it is difficult to obtain all types of variables from each grid through experiments. The objective of this paper is to investigate the degree of influence of the input parameters on the factor of safety based on a high-risk area (HRA) model to suggest the number of minimum input parameters required to obtain a reliable factor of safety. Random forest (RF) and partial dependence (PD) algorithms as well as Shapley additive explanations (SHAP) are selected to find the feature contributions in the HRA. In total, 1800 data items for each input parameter are collected; these are used as input data for the RF, PD, and SHAP algorithms. The influencing order of the independent data slightly differed, depending on the algorithm used. The input variables ranked first to fourth present a similarity. The variables ranked as important are individually applied as input data to create a deep neural network, and the predicted factor of safety is compared with the true value. An oversampling algorithm is also used to investigate the effect of the number of data on determining the influencing degree in each variable. The results demonstrate that this method could be applied to discover key parameters if obtaining all types of input variable is difficult.