The rational selection of the proportion between geological disasters (positive samples) and non-geological disasters (negative samples) holds significant importance in enhancing the precision of geological disaster susceptibility assessment and maintaining the sustainable development of the ecological environment. This paper, using Liulin County as an example, employs correlation analysis to select appropriate evaluation factors. A Random Forest (RF) model, based on GIS technology, is used for susceptibility mapping. Sample ratios of 1:1, 1:1.5, 1:3, 1:5, and 1:10 are applied. The results indicate that, through a confusion matrix test, the model's predictive performance reaches a "tipping point" at a sample ratio of 1:5. The receiver operating characteristic (ROC) curve test shows that the 1:5 model performs best. Combining the proportion of susceptibility zones and disaster points, 1:5 is identified as the most suitable ratio for assessing geological disaster susceptibility in the study area. High and very high susceptibility zones are primarily concentrated in the central and northern regions alongside roads and rivers, making these areas key focuses for disaster prevention and reduction in Liulin County. The accuracy of the model's predictions increases with a greater number of samples, but it does not continue to rise indefinitely; accuracy declines after a critical threshold is crossed. These research findings complement prior studies, promote advances in geological disaster prevention technology, and maintain geological environmental stability, all of which are crucial for the local economy's stability and sustainable development.
Keywords: Confusion matrix; Geological disaster; ROC; Random forest; Sample proportion; Sustainable development.
© 2023. The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature.