Lihao Feng, PhD
L.H. Feng1,3,4, Q.F. Wang1,2,3, D.D. Gregory4, L. Yang1, Y.Y. Niu1,3,5, H.S. Zhao1,3
1State Key Laboratory of Geological Processes and Mineral Resources, China University of Geosciences (Beijing), Beijing, China
2State Key Laboratory of Nuclear Resources and Environment, East China University of Technology, Nanchang, Jiangxi, China
3Frontiers Science Center for Deep-time Digital Earth, China University of Geosciences (Beijing), Beijing, China
4Department of Earth Sciences, University of Toronto, Toronto, Ontario, Canada
5School of Information Engineering, China University of Geosciences (Beijing), Beijing, China
The application of machine learning (ML) in mineral prospectivity mapping (MPM) has made remarkable progress in recent years. However, the current mainstream pure data-driven method often ignores the geological process, resulting in the lack of interpretability and generalization ability of the model. In order to solve this problem, this study proposes a knowledge-data dual driven prediction framework that integrates deposit system knowledge and ML. Taking the orogenic gold deposits in Guangxi as the object, the key factors such as ore-controlling structures, stratigraphic traps and geochemical anomalies were systematically integrated to construct a variety of deposit proxy models, and the algorithms such as random forest (RF), support vector machine (SVM), deep neural network (DNN) and deep forest (DF) were introduced for comparative analysis. The results show that DF model has the highest prediction accuracy and robustness, can effectively describe the non-linear relationship, and generate the prediction results consistent with the geological law. The analysis of feature importance further reveals that structural parameters are the main controlling factor, followed by geochemical and stratigraphic information. The model has identified several potential gold targets in the Youjiang Basin and Qin-Hang Belt in Guangxi, verifying its application ability. This study shows the great potential of geological knowledge embedding in improving the credibility and practicability of ML model, and provides a new technical path for integrating expert knowledge and ML in geoscience modeling in the future.