Comparison of general kernel, multiple kernel, infinite ensemble and semi-supervised support vector machines for landslide susceptibility prediction

Landslide susceptibility prediction is a key step in preventing and managing landslide hazards. As a classical supervised non-parametric machine learning model, support vector machine (SVM) has been widely used in landslide susceptibility prediction in recent years. However, most studies focus on the application of general SVM methods, or compare SVMs as benchmark methods. SVMs with different kernel functions are rarely used in this field. In this study, we apply the general SVM and its popular variants (i.e., multiple kernel learning, infinite ensemble SVM and semi-supervised SVM) to predict landslide susceptibility, and compare their prediction performance. The experimental results show that the Laplacian-SVM has the highest prediction performance (AUC = 0.8815) among SVM-based methods. SVMs with RBF kernel can achieve higher performance than SVMs with linear kernel, indicating that RBF kernel is more suitable for solving susceptibility prediction problems. Furthermore, SVM-based methods have higher sensitivity (0.8543–0.9288) than deep learning methods (0.8237–0.8271), which proves the advantage of SVMs in finding potential landslide areas.