基于BO-SVM的滑坡稳定性可解释机器学习预测方法

An explainable machine learning prediction method forlandslide stability based on BO-SVM

滑坡灾害会造成严重的生命财产安全问题,对滑坡稳定性进行准确预测对于指导滑坡灾害的防治工作具有重要意义。不平衡数据集会影响机器学习模型预测的准确性,为解决这一问题,提出一种基于SMOTE技术的BO-SVM滑坡稳定性预测方法,该方法首先采用SMOTE技术对不平衡滑坡数据集进行过采样处理,然后采用BO算法基于过采样处理后的数据集对SVM模型进行超参数优化,最后使用优化后的BO-SVM模型对测试集进行预测,其预测准确率达到95.58%。将BO-SVM模型的预测结果与RF、XGBoost和CatBoost模型的预测结果进行对比,证明BO-SVM模型的预测性能具有优越性。采用SHAP方法对SVM模型进行解释,并实施消融实验研究其输入特征对4个机器学习模型预测性能的影响。本研究解决了不平衡数据集会降低滑坡稳定性预测准确性的问题,并为滑坡的防治工作提供一定的参考。

Landslide disasters can cause serious safety of life and property.Accurate prediction of landslide stability is of great significance for guiding the prevention and control of landslide disasters.Unbalanced data sets will affect the prediction accuracy of machine learning models.In order to solve this problem,a BO-SVM landslide stability prediction method based on SMOTE technology is proposed.Firstly,the SMOTE technology is used to oversample the unbalanced landslide data set.Then,the BO algorithm is used to optimize the SVM model based on the oversampled data set.Finally,the optimized BO-SVM model is used to predict the test set.The prediction accuracy is 95.58%.The prediction results of the BO-SVM model are compared with the prediction results of the RF,XGBoost and CatBoost models to prove the superiority of the prediction performance of the BO-SVM model.The SHAP method was used to explain the SVM model,and ablation experiments were performed to study the influence of input features on the prediction performance of the four machine learning models.This study solves the problem that the unbalanced data set will reduce the accuracy of landslide stability prediction,and can provide some reference for the prevention and control of landslides.