Feature Engineering of Geohazard Susceptibility Analysis Based on the Random Forest Algorithm: Taking Tianshui City, Gansu Province, as an Example

In this paper, Feature Engineering (FE) was applied to Landslide Susceptibility Mapping (LSM), while the most suitable conditioning feature dataset and analysis method were tested analyzed. Tianshui city taken as study area, three types of geohazard (collapse, landslide, unstable slopes) used, a total twenty-three features generated; two dimensionless methods (normalization standardization) afterward. Four Random-Forest-based (RF-based) selection using different indicators (Gini Impurity, GI; Out Bag Accuracy, OOBA) proposed separately. The LSMs four models carried out under guidance results FE, namely Classification Regression Tree (CART), Random Forest (RF), Logistic (LR), Support Vector Machine for (SVC). For enhancement, standardization had significant advantages over normalization. All RF-based proven effective, lifting AUC by 0.01~0.02. RF model achieved highest LSM accuracies, respectively, 0.949 (landslide), 0.957, (unstable slopes), improved 0.008 0.005 (collapse), 0.013 slopes). This proved that FE helped improve can help decide dominant factors regional geohazards.