17:15 〜 18:45
[HDS08-P21] Research on a classification model of loess seismic landslides based on random forest in the Haiyuan region
キーワード：Loess seismic landslide, Landslide classification, Random Forest algorithm, Model building
The 1920 Haiyuan earthquake in Ningxia induced many loess landslides that still significantly impact the safety of local people. Loess seismic landslides can be mainly divided into three types: the shear-sliding type, seismic-collapsibility type, and liquefaction type, with substantial differences in the deformation characteristics, damage mechanism, and hazard risk. The identification of landslide type can provide a basis for assessing, preventing, and mitigating geological hazards. A total of 1163 landslides (including 141 liquefaction landslides and 775 shear-sliding landslides) were interpreted. The distribution, topography, motion, mechanism, and relevant parameters of the landslides were systematically analyzed. A model of loess seismic landslides classification based on the random forest was proposed. A sample dataset with 21 input parameters of landslide characteristics and one output parameter of landslide type was constructed. Based on the random forest algorithm, the sample dataset was trained with fivefold cross-validation. The parameters in the model were adjusted by a Bayesian optimization algorithm. The model’s generalization ability was verified adopting the receiver operating characteristic (ROC) curves and AUC (the area under the ROC curve) values. The results showed that the classification model has high accuracy and generalization ability (with an accuracy of 92.3% and an AUC value of 0.978). When the landslide classification is carried out for a specific region, the characteristic attributes can be input according to the training samples. The classification model is invoked to set the threshold value of 0.5 and output probability. The landslide type can be determined by the probability subsection according to the output probability P.