Abstract
The urban agglomeration in central Guizhou is located in a crustal deformation area caused by tectonic uplift between the Mesozoic orogenic belt of East Asia and the Alpine-Tethys Cenozoic orogenic belt, with high mountains, steep slopes, fractured rock masses and a fragile ecological environment; this area is the most affected by landslides in Guizhou Province, China. From 2011 to 2023, there were a total of 648 medium and large landslide disasters, resulting in 143 deaths and a direct economic loss of 1.191 billion yuan. Therefore, this study selected 12 indicators from the topography, geological structure, and external inducing factors, and conducted factor collinearity analysis using the variance expansion coefficient to construct a landslide hazard assessment index system. The statistical analysis model was combined with a variety of machine learning models, and the selection of negative sample points was restricted in various ways to improve training data accuracy and enable machine learning model predictions with sufficiently supervised prerequisites. The accuracy of the model was validated by ROC curve analysis. The AUC values of the SVM, DNN, and bagging models were all greater than 0.85, indicating that the results were credible. However, the overall accuracy was SVM > DNN > Bagging; that is, SVM was more suitable for landslide hazard assessment of the urban agglomeration in central Guizhou. Finally, field surveys were used to validate multiple sites with historical landslides in extremely high-hazard areas and analyse their development characteristics. The evaluation results can provide strong guidance for engineering design, construction and disaster prevention decision-making of urban agglomeration in central Guizhou.