Study On Metallogenic Prediction Based On The Fusion Model Of GBDT And SVM

In the field of mineralization prediction research,geological data has a large amount of data and rich types,which shows the advantages of machine learning algorithms.The use of machine learning algorithms for mineral prediction research is gradually replacing traditional methods of prospecting.However,due to data problems or the limitations of each algorithm,a single algorithm has its own advantages and disadvantages in practical application.How to make up for the deficiencies of data or algorithms,and make better use of the advantages of machine learning,is worthy of research and exploration.Therefore,this paper improves the application model of machine learning classic algorithm support vector machine(SVM)in the field of metallogenic prediction,The geochemical data,geological structure data,stratum and rock mass data of Eastern Tianshan research area are used to build a gradient boosting decision tree(GBDT)and support vector machine(SVM)fusion model algorithm was used to predict the mineralization of the eastern Tianshan Mountains in Xinjiang.The main research contents and results are as follows:(1)Sort out the mineral control elements and perform data preprocessing.Systematically collected data from the research area to understand the mineralization pattern of the research area;preliminary determination of ore-controlling elements from multivariate data types,and data processing of ore-controlling elements using GIS technology to construct a machine learning model training dataset(2)Construction of feature combinations.The selection and processing of features in the field of mineralization prediction are mostly based on expert experience and there is a large uncertainty.Therefore,based on the traditional feature selection,this paper proposes the GBDT-SVM model.The GBDT algorithm is used to construct the feature combination,and then the feature combination is used as a new feature to build a SVM classification model.This makes up for the deficiency of SVM algorithm in feature selection.(3)Improved PSO algorithm.One of the core problems in the SVM algorithm is the selection of parameters.In this paper,the particle swarm optimization(PSO)algorithm is used for automatic optimization of the kernel function parameter combination,which is easy to fall into the local minimum in the later stage of optimization.The local minimum problem was improved to improve the classification accuracy of the model.(4)Using the GBDT-SVM model proposed in this paper,and the random forest(RF)model,the SVM model,the GBDT model,the metallogenic prediction of Xinjiang's eastern Tianshan district was carried out respectively,and the prediction results are compared and analyzed.Through the intuitive comparison of the prediction results and the comparison based on the model accuracy(ACC)and the area under the ROC(AUC)indicator,it is concluded that: the GBDT-SVM fusion model has good distinguishing ability between mines and non-mines compared to the single prediction model,the overall classification accuracy is the highest,and the prediction effect is the best.