A comparative approach of support vector machine kernel functions for GIS-based landslide susceptibility mapping

Landslides are among the most destructive natural hazards with severe socio-economic ramifications all around the world. Understanding the critical combination of geoenvironmental factors involved in the occurrence of landslides can mitigate the adverse impacts ascribed to them. Among the several scenarios for studying and investigating this phenomenon, landslide susceptibility mapping (LSM) is the most prominent method. Applying the machine learning (ML) algorithms integrated with the geographic information systems (GIS) has become a trending means for accurate and rapid landslide mapping practices in the scientific community. Support vector machine (SVM) has been the most commonly applied ML algorithm for LSM in recent years. The current study aims to implement different SVM kernel functions including polynomial kernel function (PKF) (degree 1 to 5), radial basis function (RBF), sigmoid, and linear kernels, for a GIS-based LSM over the Tabriz Basin (TB). To this end, a total number of 9 conditioning parameters being involved in the occurrence of the landslide events were determined and utilized. The LSM maps of the TB were generated based on the different SVM kernels and were statistically validated according to the landslide inventory. The findings revealed that the polynomial-degree-2 (PKF-2) model (AUC = 0.9688) outperforms the rest of the utilized kernels. According to the SLM map generated through PKF-2, the northernmost parts of the TB are extremely susceptible to slope failures than the rest; therefore, the developmental policies over these parts have to be taken into account with privileged priority to hinder any humanitarian as well as environmental catastrophes.