Geochemical Anomaly Extraction And Mineralization Prediction Based On Fractal Theory And Machine Learning

The Heilongjiang Jianbiannongchang area is located at the confluence of the Great Xing’an Range and the Lesser Xing’an Range and has undergone a long and complex superimposed tectonic zone of magmatic and tectonic evolution.The area is one of the most complex and intense areas of tectonic-magmatic evolution in the northeast,forming a variety of types of non-ferrous minerals and precious metals with high resource development value.Further work on resource quantification and prospect prediction for the area is therefore necessary.This paper takes the quantitative mineral resource prediction and evaluation system as a guide and uses 1:50,000 soil geochemical data from the Jianbiannongchang area of the Heilongjiang River as the basis for collecting sufficient geological data and previous research results.It introduces compositional data analysis methods and uses the R language,Python,Geo DAS,and Arc GIS platforms.The nonlinear fractal methods(such as the local singularity analysis method,concentration-area fractal method,and spectrum-area fractal method)and machine learning methods(random forest method and support vector machine method)are used for geochemical anomaly identification and mineralization prediction.The main results and progress achieved are as follows:1.Isometric log-ratio transformations and exploratory data analysis were conducted on major mineralization indicators,such as Au,Cu,Pb,Zn,and Sb,in 1:50,000 soil geochemical samples.The relationship between elemental assemblages and mineralization was investigated using robust principal component analysis based on ILR transformations.The results were combined with the interpretation of the principal component double-labeled map to clarify the significance of elemental assemblages on mineralization.The obtained PC1 and PC2 principal component information reflected mineralization basic anomalies dominated by Cu and Au,respectively.2.The concentration-area(C-A)fractal method,spectrum-area(S-A)fractal method,and local singularity analysis method(LSA)were used to investigate and quantify the geochemical distribution patterns of elemental assemblages and mineralogical anomalies defined by robust principal component analysis.The study and quantitative analysis highlight the hidden weakly slow anomalies in the deep and low background areas of the study area,which can effectively indicate the distribution of hidden deposits.3.Geochemical mineralization models were developed with the help of random forest(RF)and support vector machine(SVM)models in machine learning methods.The model performance of the random forest and support vector machine was assessed by ROC curves,high potential spaces associated with geochemical mineralization were identified,and the effectiveness of the random forest and support vector machine models in predicting mineralization from geochemical data in the Jianbiannongchang area of Heilongjiang was explored.4.The results of the obtained fractal theory and machine learning methods were compared and analyzed,and the results showed that the SVM geochemical mineralization prediction model established in the Jianbiannongchang area of Heilongjiang was the best.Therefore,this study will be carried out based on the SVM model to define the prospective mineralization zones in the Jianbiannongchang area and six prospective mineralization zones have been defined.The paper has 34 figures,5 tables,and 135 references.