Enhancing Spectral Clustering Performance Using Self-Supervised Support Vector Machines for Regional Landslide Risk Assessment Visualization: A Case Study in Han-Yuan County, Ya’an City

In this paper, A novel application, self-supervised support vector machine learning, is proposed to enhance the performance of an unsupervised spectral clustering model in regional landslide risk assessment visualization. Taking Han-Yuan County, Ya’an City, as the case study, we utilized spectral clustering to train the sample set and attained the optimal kernel function parameter value of 0.03, which was then applied to the total sample for classification prediction and an AUC value of 0.819 was achieved. To further improve the model performance, we applied self-supervised learning on the basis of the spectral clustering. After introducing this approach, the model performance was substantially improved, with an AUC value of 0.954, representing an increase of 13.5% compared to the unsupervised spectral clustering, and the calculation time was increased by only 0.017 s, which was negligible. The improved algorithm also produced a landslide susceptibility risk evaluation map for Han-Yuan County, Ya’an City, indicating that the high-risk area and the extremely high-risk area occupied 27.05% of the entire landslide, but accounted for 83% of the landslides, confirming the efficacy of this method in terms of enhancing the accuracy of spectral clustering.