High-grade zone estimation using the SVM and BPPN algorithms in Chah Firuzeh porphyry copper deposit

Mosatafaei, Kamran; Hezarkhani, Ardeshir; Maleki, Shahoo; Zamani Ahmad Mahmoudi, Mohammad; Khalilidermani, Mitra; Knez, Dariusz

doi:10.22059/ijmge.2026.401926.595303

High-grade zone estimation using the SVM and BPPN algorithms in Chah Firuzeh porphyry copper deposit

Articles in Press

Document Type : Research Paper

Authors

¹ Assistant Professor, Department of Mining Engineering, Faculty of Engineering, University of Kurdistan, Sanandaj, Iran

² Faculty of Mining Engineering, Amirkabir University of Technology, Tehran, Iran

³ Department of Drilling and Geo-engineering, Faculty of Drilling, Oil and Gas, AGH University of Krakow, Krakow, Poland

10.22059/ijmge.2026.401926.595303

Abstract

In the Chah Firuzeh porphyry copper deposit, a number of thirteen coring boreholes were drilled to evaluate the copper grades in the anomaly. While twelve of the boreholes intersected only the low- and medium-grade copper zones, the borehole CHF06 reached a high-grade zone. As such a high-grade zone is economically invaluable in the financial perspective of the mine, the corresponding copper grades must be estimated precisely. The primary goal of the current study is to estimate the copper grades in such high-profit zone using three artificial intelligence (AI) techniques: Support Vector Machine (SVM) and Back Propagation Neural Network (BPNN). Due to the porphyry nature of deposit, no clear relation was found between the copper grades of the borehole CHF06 and the rest. To address this issue, the Genetic Algorithm-Artificial Neural Network (GA-ANN) and Principal Component Analysis (PCA) algorithms were utilized to choose the best input dataset for those three AI techniques. Both the GA-ANN and PCA algorithms detected that the copper grades of the boreholes CHF05, CHF21, CHF24, and CHF26 are the most appropriate input data to be imported into the SVM and BPNN models. After grade estimation, the R-square (R²) of the SVM and BPNN, techniques were obtained as 0.98 and 0.72 respectively. Moreover, further analysis uncovered that the SVM model has the least sensitivity to the ratio of training data to testing data. Hence, the SVM approach was recognized as the most reliable AI technique to accurately solve the complex resource estimation problems in mining projects. This key finding implies that a SVM estimator can be applied not only for the uniform-mineralization ores but also for the deposits exhibiting a highly inconsistent grade-trend in their structures.

Keywords

Main Subjects