Landslide susceptibility assessment via the information value-coupled machine learning models

Collapses and landslides are frequent in the southern mountainous areas of the economic zone on the northern slopes of the Tianshan Mountains in Xinjiang, and an accurate assessment of susceptibility can effectively avoid potential risks, which is crucial for the prevention and control of geological hazards. To obtain precise and reliable references for the prevention of landslide, a total of 10 landslide conditioning factors (e.g., elevation, slope degree, slope aspect, curvature, relief, engineering geological lithology, landform types, land use, distance to rivers, as well as distance to roads) were selected for the multicollinearity analysis. The evaluation index system was established in the present research to assess the landslide susceptibility with the combination of traditional statistical methods and machine learning models. Both the information value-maximum entropy coupled model (I-MaxEnt) and the information value-logistic regression coupled model (I-LR) were proposed to assess landslide susceptibility in the Tianshan northern slope economic belt after the detailed evaluation on the information value model (I), logistic regression model (LR), and maximum entropy model (MaxEnt). Comparative discussions on the receiver operating characteristic (ROC)curves revealed that the area under the curve(AUC) values of the I-MaxEnt and I-LR coupled models were 0.907 and 0.941, respectively, indicating the superior accuracy of the I-LR model. Furthermore, the results obtained from the I-LR model were more consistent with the actual situation as verified by field validation. That is, the I-LR model is much more suitable in assessing the landslide susceptibility in the given research region attributed to its high accuracy and reliability.The results of this research provide a reliable basis for disaster prevention and mitigation in this study area.