Evaluation Model Of Random Forest And Support Vector Machine For Landslide Prone Along Mountain Road

The mountainous area in southwestern China is relatively extensive.Human life and production activities are all carried out in mountainous areas.However,due to transportation problems,economic conditions are restricted.In order to develop the economic development in the southwest,more and more road traffic and other projects have been built.Most of the roads in the mountainous areas are built in the gentle river valley.However,due to the complex terrain,some roads need to be excavated.The construction of roads involves the excavation and grading of roadbeds,which will lead to the development of more fissures along the road slope and reduced stability,resulting in landslides and other geological disasters.Mountain roads are frequent areas of landslides,which can cause road damage and threaten human life and property.Research on landslides along the road can provide a basis for the prevention and control of road landslides.This research area is mainly in the northeast of Chongqing.Chengkou County is selected as the research area,and Yunyang County is the verification area.The study is divided into slope units.Random forests,support vector machines and GIS systems are used to facilitate the occurrence of landslides along mountain roads.Evaluation.The main research contents and results are as follows:(1)Study the topography,geological structure,stratigraphic lithology and meteorological and hydrological conditions of northeast Chongqing in the study area,and study the road and landslide data of Chengkou County and Yunyang County.The mechanism of landslide occurrence is analyzed through the overview of the study area,and the evaluation factors of the study are selected accordingly.The topography and landform factors are selected as elevation,slope,aspect,plane curvature,profile curvature,comprehensive curvature(surface analysis),slope position,terrain roughness,micro-landform,topography humidity index(TWI),water flow index(SPI),sediment transport Index(STI),geological structure factor selects the type of rupture,lithology,distance to fault,geological environmental factor selects NDVI,distance to river,human engineering activity selects the distance to the road this time,and the inducing factor selects average years of rainfall.(2)Divide the slope units in Chengkou County and Yunyang County,and select the slope units in the 3 km left and right along the road as the study area.Using the slope unit to extract 19 evaluation factors,construct a geospatial database.(3)Random forest and support vector machine are combined with geographic information system to construct a landslide susceptibility evaluation model along mountain roads,and the accuracy of the two models is evaluated using ROC curve and confusion matrix.The ROC values of the training data,verification data and overall data of the model established by the random forest algorithm are 0.999?0.802 and 0.951 respectively,and the recall,precision and accuracy in the confusion matrix are 0.933?0.929 and 0.876,respectively.The ROC values of the training data,verification data and overall data of the model built by the support vector machine algorithm are 1.000?0.770 and 0.954,respectively,and the recall,precision and accuracy in the confusion matrix are 0.906?0.947 and 0.834 respectively.The accuracy and accuracy of these two landslide prediction models are relatively high.(4)Simulate the two evaluation models on the entire research area of Chengkou County and Yunyang County,and verify the results of the simulation using mathematical statistics and ROC curves.In mathematical statistics,the simulation results of the random forest algorithm and the support vector machine algorithm in Chengkou and Yunyang show that there are more slope units in areas with low susceptibility,and fewer slope units in areas with high susceptibility,but There are many landslides,and the density of landslides increases as the susceptibility level increases.The AUC values of the simulation results of the random forest algorithm and the support vector machine algorithm of Chengkou County are 0.865 and 0.821,respectively,and the AUC values of the simulation results of the random forest algorithm and the support vector machine algorithm of Yunyang County are: 0.824 and0.812.Both the random forest model and the support vector machine established landslide susceptibility evaluation model have good accuracy,and the evaluation model constructed by random forest has a better effect.