Application of Supervised Machine Learning Inversion in the Estimation of Iron Ore Grade from Geophysical Data: Comparative Investigation of GB, RF and SVM Algorithms

Simorgh, Mohsen; Alimoradi, Andisheh; Hemmati Ahooi, Hamidreza; Salsabili, Mohammad; Fathi, Mahdi; Ghasemi Tabar, Hassanreza; Rezakhani, Parisa

doi:10.22077/jgm.2024.7358.1018

Application of Supervised Machine Learning Inversion in the Estimation of Iron Ore Grade from Geophysical Data: Comparative Investigation of GB, RF and SVM Algorithms

Document Type : Original Article

Authors

Mohsen Simorgh ¹

Andisheh Alimoradi ²

Hamidreza Hemmati Ahooi ¹

Mohammad Salsabili ³

Mahdi Fathi ⁴

Hassanreza Ghasemi Tabar ⁵

Parisa Rezakhani ¹

¹ Department of Mining Engineering, Imam Khomeini International University

² Department of Mining and Petroleum Engineering, Imam Khomeini International University

³ Department des Sciences Appliquees, Universite du Quebec a Chicoutimi

⁴ Kavoshgaran Consulting Engineers

⁵ Department of Mining, Petroleum & Geophysics, Shahrood University of Technology

10.22077/jgm.2024.7358.1018

Abstract

Magnetometry is one of the geophysical methods used to explore metal deposits, especially iron deposits and magnetite minerals. The two-dimensional model resulting from the magnetometric operation cannot estimate the grade in the depth of the deposit, so in this article, the attempt is made by using the magnetic outputs obtained after the magnetic survey operation and the two-dimensional model designed with the help of the data extracted from the borehole which is available in the studied area, and combining this information and obtaining relationships between them with the help of artificial intelligence, a three-dimensional numerical model can be obtained that can be generalized to other points that lack depth data. This method will be a new approach to numerical simulation in the field of investigation of mineral masses.

Finally, in the studied area of the Sechahoon deposit in central Iran, high precision was achieved in the ratio of zero iron grade data in the methods of Gradient Boosting and Random Forest. Also, the results of these two algorithms showed that the Maximum Mean Square Error (MSE) and Mean Absolute Error (MAE) in the training data are 0.007 and 0.05, respectively, and in the test data are 0.03 and 0.11, respectively, which these parameters reached the maximum of 0.03 and 0.1 in the inspection of validation boreholes.

Keywords

Magnetometry

Gradient Boosting

Random Forest

Support Vector Machines (SVMs)

Three-dimensional Modeling

Iron Deposit