基于FR的多种机器学习模型在地质灾害易发性评价中的对比分析

Comparative Analysis of Various Machine Learning Models in the Evaluation of Geological Disaster Susceptibility

选取建模效果较好的3种模型,支持向量机模型、随机森林模型和人工神经网络模型,以福建福鼎为研究区进行地质灾害易发性分析,将3种模型建模结果通过ROC曲线和频度比法进行对比,选择最优福鼎地质灾害易发性分析结果,为该地区防灾减灾提供有利指导。结果表明:3种模型均能有效地预测出福鼎地质灾害易发性,是基于该区域可靠性较高的预测模型;支持向量机模型的准确率(86.7%)高于人工神经网络(84.7%)和随机森林(79.7%)的准确率;支持向量机模型的频度比分析值更具梯度,是最适合该地区的易发性分析模型。

Three models with good modeling effect,support vector machine model,random forest model and artificial neural network model,are selected,which are of great significance in geological hazard susceptibility analysis.Taking Fuding,Fujian Province as the study area,the geological disaster susceptibility analysis is carried out,and the modeling results of the three models are compared through ROC curve and frequency ratio method,selecting the optimal analysis results of geological disaster susceptibility in Fuding.The results show that:the three models can predict effectively the susceptibility of geological disasters in Fuding,which is based on the prediction model with high reliability in this region.The accuracy of SVM(86.7%)is higher than that of ANN(84.7%)and RF(79.7%).The frequency ratio of SVM is better than the two,so that SVM is the most suitable susceptibility analysis model in this region.