Evaluation of Landslide Susceptibility Using Machine Learning Based on Information Value Sampling Method

In landslide susceptibility prediction using machine learning (ML) models, the selection of appropriate negative samples significantly impacts the accuracy of the predictions. Many existing studies tend to randomly choose negative samples from the entire research area or specific attribute areas such as low slopes, leading to inaccuracies and biases that decrease the reliability of susceptibility mapping. To address this issue, this paper proposes an ML model based on the coupled information value (IV) method for susceptibility modeling. In this approach, the attribute values of environmental factors are transformed into information values, representing their contribution to landslides, using the information quantity method. Very low and low prone areas are then identified based on these information values, and random points are selected from these areas as negative samples for training and verifying subsequent machine learning models, which include the Random Forest (RF) model and Support Vector Machine (SVM) model.The study focuses on the Three Gorges reservoir area in the Zigui-Badong area as an example. The comparative analysis of models trained using traditional sampling methods reveals that the RF model and SVM model, which were sampled using the information value method, achieve accuracies of 95.33% and 91.13%, respectively. In contrast, models trained using traditional sampling methods achieve accuracies of 90.43% and 88.69%, respectively.