耦合PCA-SVM算法的金矿矿床规模预测分析研究

Identification of ore deposit types is an important part of mineral exploration. Traditional methods for predicting deposit size are time-consuming, laborious and costly. In order to improve prospecting efficiency and accuracy and reveal potential relation between chemical composition and the size of gold mineralization, we propose here an integrated approach using the Principal Component Analysis (PCA) and Support Vector Machine (SVM) algorithms. In this approach, we first extract the major features of samples using PCA, and we then train a set of SVM classifiers by these features to predict deposit sizes. In this study, we collected and analyzed 3812 gold mine samples from Beishan, Gansu region to establish a PCA-SVM model with the training accuracy of 92.3% and the test accuracy of 88.7%, which were 14.3% and 17.1% higher, respectively, than using SVM. We demonstrated that the PCA-SVM method not only can eliminate subjective factors, but also can improve the accuracy of identifying ore deposits as well as prospecting efficiency, thus to provide reliable support for decision making.