GIS-based mineral prospectivity mapping using machine learning methods: A case study from Duobaoshan ore district, northeastern China

Abstract

• Machine learning models integrated with GIS achieve high mineral prospectivity mapping accuracy. • RBFLN, SVM, and RF models demonstrate superior predictive performance. • Comparison with WOE highlights the advantages of machine learning in complex geological settings. • Workflow provides a systematic approach for delineating mineral exploration targets. Exploration and development of mineral resources rely on the accurate prediction of mineral distribution patterns. The effective integration of multi-source geo-datasets and the improvement of prediction accuracy have become key challenges in Mineral Prospectivity Mapping (MPM). Traditional prediction methods often inadequate when facing with complex geological setting and non-linear correlations. Recently, Geographic Information System (GIS) technology has emerged as a powerful tool for integrating multi-source spatial data, while machine learning algorithms, with their advantages in handling high-dimensional and complex data, have become essential for enhancing the efficiency and accuracy of MPM. The research focuses on the Duobaoshan ore district in northeastern China, employing GIS-based machine learning methods to generate a mineral prospectivity mapping and enhance the accuracy of target area delineation for potential mineral deposits. The workflow is as follows: (i) establishing a metallogenic model by investigate the regional mineral system, identifying ore-controlling factors, and generating mappable exploration criteria; (ii) constructing a exploration model based on comprehensive ore-controlling factors and exploration indicators to provide input data for predictive evaluation; (iii) extracting spatial proxies to develop predictive models; (iv) applying Radial Basis Functions Link Network (RBFLN), Support Vector Machine (SVM), and Random Forest (RF) algorithms to capture the complex nonlinear relationships between known deposits and evidence layers; (v) evaluating the prediction model results using the Receiver Operating Characteristic (ROC) curve, with Area Under the Curve (AUC) values for RBFLN, SVM, and RF at 0.984, 0.920, and 0.858, respectively, indicating high predictive performance across all three models; and (vi) Additionally, this study incorporates a comparison with the results of the Weight of evidence(WOE) method, which achieved an AUC value of 0.717, further validating the superior performance of GIS-based MPM in identifying prospective areas using machine learning. Based on the results from the three models, a prospectivity map was generated, delineating high, moderate, and low prospectivity areas including two Class I targets and two Class II targets, thus providing a foundation for further mineral exploration in the area.

Topics & Concepts