A Study On The Impact Of Support Vector Machines On The Accuracy Of Landslide Susceptibility Evaluation Based On Multiple Algorithm Tuning

Landslide is a more frequent geological disaster,which seriously affects all kinds of engineering construction and human activities.Therefore,in various engineering constructions,it becomes a very important work to avoid the high landslide susceptibility area and provide reliable landslide susceptibility zoning map for the siting of various projects.In recent years,there are more and more studies on the landslide susceptibility of a certain area using machine learning methods,but how to obtain a more accurate landslide susceptibility zoning map is still a problem worth studying.In this study,the range along a 60 Km-long section of oil pipeline in Kunming,Yunnan Province was selected as the study area,and the information of 141 landslide points in the study area was obtained through field survey and data collection,while lithology,elevation,slope direction,slope,SPI,TWI,average annual rainfall(2017-2021),distance to the road,and terrain roughness were selected as the conditions for landslide susceptibility study in the study area factors.Firstly,the magnitude of the influence of the condition factors on landslides was quantified using the informative method,and then the parameters were tuned using genetic algorithm(GA),particle swarm algorithm(PSO),and bat algorithm(BA),respectively,and then the study area was analyzed for landslide susceptibility using SVM,and finally the untuned SVM with accuracy,precision,recall,subject work characteristic(ROC)curve,and Lift elevation model and GA,PSO,and BA tuned SVM model performance were examined.The results of the study showed that:1.The results obtained from the four evaluation models are consistent with the actual situation and have certain implications for realistic engineering practice.Among the evaluation results of the four machine learning models,most of the landslide sites in the study area are developed in areas with high and very high susceptibility ratings.Among them,the BA-SVM model with very high susceptibility rating has significantly more information than the other three models,indicating that its results are more consistent with the actual situation in the study area.2.This study used a variety of metrics to test the model results,and the results showed that the SVM,GA-SVM,PSO-SVM and BA-SVM models all performed relatively well,with the BA-SVM model showing the best performance with higher reliability and stability.3.This study compares the application results of different optimization algorithms and finds that the BA algorithm is more effective in the SVM model compared to the GA and PSO algorithms.4.It is feasible to use machine learning methods for landslide susceptibility evaluation,and all four models,SVM,GA-SVM,PSO-SVM and BA-SVM,have certain accuracy and prediction ability.Among them,the SVM model optimized by BA algorithm is better than the other three models in all aspects.Therefore,the BA-SVM model is the most suitable landslide susceptibility evaluation model for the study area in this study.This study is of great practical significance for the evaluation of landslide susceptibility along the oil pipeline in the study area,which can provide scientific basis and technical support for the safety of oil and gas transmission and also provide useful reference and reference for the study of similar problems in other areas.