Research On Landslide Geological Disaster Prediction Model Based On Machine Learning

China has a vast territory and complex natural environment and climate conditions,which lead to frequent occurrence of geological disasters.At the same time,geological disasters had caused great harm to the safety of people’s lives and property and social economic development.According to the survey data,landslides account for more than 70%of all geological disasters,which is difficult to control and widely distributed.Therefore,how to effectively predict landslide geological hazards has a great significance to provide scientific basis for the prevention and control of landslide geological hazards.In this paper,Ningqiang County of Shaanxi Province was taken as the research area to study the landslide geological disaster prediction model.The specific research content and results were as follows:(1)Firstly,the main influencing factors of landslide geological disasters are selected by analyzing geological environment and landslide formation conditions in the study area.Then,Landsat8 remote sensing images and elevation digital model were analyzed based on GIS technology,and 18 landslide impact factors were extracted.The missing values and dimensional differences in the data were processed by random forest and Min-max method.Finally,principal component analysis and Pearson correlation coefficient method were used to obtain 10 landslide influence factors with large weight an d weak linear correlation..(2)Four types of landslide disaster prediction models were built based by using vector machine,logistic regression,random forest,and Adaboost machine learning algorithm,respectively.Notably,bayesian optimization algorithm was used to optimize the important hyperparameters of the model.The results show that except for the multilayer perceptron neural network model,the accuracy,F1 score and AUC of the other four models have reached more than 0.8,which lays a foundation for the next step of model fusion.(3)Owing to the traditional stacking model fusion algorithm ignores the correlation between the characteristic variables and the output values and the differences in the learning effect of the basic learning machine,this paper proposes an improved stacking algorithm which combines the secondary learning layer with the original features and precision feedback weighting.Based on this,support vector machine,logistic regression,random forest and AdaBoost model are fused to build a fusion landslide geological disaster prediction model.The experimental results showed that,compared with other models,the accuracy,F1 score which verifies the effectiveness of the improved stacking algorithm and its applicability to the landslide geological hazard prediction in the study area.(4)Based on the integrated landslide geological disaster prediction model,a landslide geological disaster prediction system integrating meteorological monitoring,environmental display,landslide prediction and data management functions was developed.The development of the system improved the timeliness of meteorological monitoring,the intuitiveness of environmental display,the scientificity of landslide prediction and the convenience of data management in the study area.