Deposit type discrimination based on trace elements in sphalerite
Yu-Miao Meng,Xiao-Wen Huang,Ruizhong Hu,Georges Beaudoin,Mei-Fu Zhou,Songning Meng
DOI: https://doi.org/10.1016/j.oregeorev.2024.105887
IF: 3.3
2024-01-15
Ore Geology Reviews
Abstract:Sphalerite can form in different geological environments and is the main ore mineral of most Pb-Zn deposits. Trace elements in sphalerite combined with machine learning (ML) methods have been used for deposit type discrimination. However, the performance of different multivariate and ML methods is not systematically compared. Here more than four thousand laser ablation inductively coupled plasma mass spectrometry data from more than one hundred deposits were compiled and investigated by different multivariate and ML methods. Unsupervised multivariate and ML methods including principal component analysis (PCA) and t -distributed stochastic neighbor embedding ( t -SNE), and supervised multivariate and ML methods including partial least squares-discriminant analysis (PLS-DA), logistic regression (LR), support vector machine (SVM), K-nearest neighbors (KNN), random forest (RF), naïve Bayes (NB), and neural network (NN) are used to classify deposit types. The models are constructed for five deposit types, Mississippi Valley-type (MVT), Volcanogenic Massive Sulfide (VMS), sedimentary exhalative (SEDEX), Epithermal, and Skarn deposits, based on twelve elements in sphalerite, Fe, Pb, Mn, Co, Cu, Ga, Ge, Ag, Cd, In, Sn, and Sb. The PCA and PLS-DA methods show similar clustering results, where Epithermal, MVT, SEDEX, and Skarn deposits can be separated from other deposit types. The t -SNE method is only efficient in discriminating MVT, SEDEX, and Epithermal deposits. The KNN, RF, and NN methods show better performance than LR, SVM, and NB in terms of average accuracy and precision. The PCA, t -SNE, and PLS-DA methods have advantages in data visualization, whereas KNN, NN, and RF methods are more suitable for classification tasks. Feature importance analyses of the PLS-DA and RF methods show that the most important discriminant elements in the classification are Ge, Mn, In, Sn, Ga and Sb. The models based on PCA, PLS-DA, KNN, NN, and RF are used to predict sphalerite samples with uncertain origins. All the models give similar classification results for deposit types, which are consistent with the ore-forming geological setting. The multivariate and ML models, including discrimination diagrams of PCA and PLS-DA, can be used as the first screening for the type of Pb-Zn mineralization in mineral exploration.
mineralogy,geology,mining & mineral processing