Porphyry Copper Prospectivity using Machine Learning Methods: A Case Study of the Shahr-e-Babak Prospecting Area, South Eastern Iran

doi:10.21203/rs.3.rs-3227701/v1

Mineral prospectivity modelling (MPM) is an essential step in reducing cost and time at the reconnaissance stage of mineral exploration. In this paper, the MPM was conducted in the Shahr-e-Babak study area for porphyry copper prospectivity. For achieving this goal, the evidential layers, including geology, remote sensing, airborne geophysics, geochemistry, and elevation model, were used as the input of training models. Four machine learning methods, consisting of multilayer perceptron (MLP), Adaptive neuro fuzzy inference system (ANFIS), random forest (RF), and generalized regression neural network (GRNN), were used to generate the models. Then, the fractal method and the prediction area plot were applied to evaluate the models. The models were divided into low potential, moderate potential, and high potential zones. The effective weight of evidential layers was extracted using the P-A plot method. The weight of Cu anomalies, phyllic, argillic and iron oxide alterations, elevation data, PC1 geochemical anomalies, magnetic anomalies, and subvolcanic bodies were 0.71, 0.62, 0.49, 0.4, 0.32, 0.25, 0.25 and -0.49, respectively. In the next step, the weights were extracted for MLP, ANFIS, RF, and GRNN as 0.85, 0.78, 1.26, and 0.76, respectively. The statistical correlation coefficients between argillic, phyllic, and iron oxide alterations were calculated. In the final step, an integrated model was generated using machine learning methods. Then, the integrated model was divided into low, moderate, and high potential zones based on the fractal method. Favorable areas are located in the western and eastern parts of the study area based on the integrated model.

Machine learning

Mineral prospectivity

Shahr-e- Babak

porphyry

Iran

The purpose of mineral prospectivity mapping is detecting new areas that have high potential for the occurrence of mineral deposits. There are two approaches to detect such areas: knowledge- and data-driven methods. While the knowledge-driven methods apply the knowledge and experience of geoscientists, the data-driven methods use the relationship between geospatial data and known occurrences in the study area.

In order to solve an MPM problem, it is necessary to collect, analyze, and integrate certain geospatial data. These data include geological, geophysical, geochemical, and remote sensing layers. In the final step, the association between known mineralization points and target areas should be surveyed and evaluated (Ghezelbash et al. 2023). Knowledge-driven methods are appropriate for poorly explored areas, whereas data-driven methods are appropriate for properly explored areas. Nowadays, Artificial Intelligence (AI) -based approaches, such as machine learning and deep learning methods, have proved to be highly effective in geoscience studies. For instance, complicated relationships between geospatial data and known mineralized points can be extracted using AI (Daviran et al. 2022).The application of machine learning methods saves time and budget in mineral exploration projects. Also, the models produced using AI-based methods are highly accurate and reliable.

An MPM method can be considered as a classification problem, for which the deposit points in the study area are considered as class one and non-deposit points as class zero (Daviran et al. 2021). The classification method in an MPM problem is carried out using training data, which are the deposit and non-deposit points. After finding the relationship between geodatabase and training points (deposit and non-deposit points), the relationship is generalized to the study area.

In recent years, different data-driven methods have been used for mineral prospectivity modelling. Random forest, support vector machine, artificial neural network, and adaptive neuro-fuzzy are some of the commonly used algorithms for this purpose. Although these algorithms were first developed by computer scientists, they have been widely used in other scientific studies, especially in geosciences.

In this paper, four machine learning methods, including artificial neural network, adaptive neuro fuzzy, generalized neural network, and random forest have been applied to the Share-Babak study area which is located on the Urumia-Dokhtar Magmatic zone in Iran. Afterward, these algorithms were compared and combined, based on which the final MPM models of the Shahr-e-Babak study area was constructed.

Study area and data used

Geology

This paper studies a part of the Urumieh–Dokhtar Magmatic Arc (UDMA) (Fig. 1), which is located on the Alpine–Himalayan orogenic belt. This belt resulted from the closure of the Neotethyan Ocean between Arabia and Eurasia. The UDMA, which is considered to be one of the most important Cu-bearing regions of the world, contains three giant and ten intermediate porphyry Cu ± Mo ± Au deposits with a total reserve of > 40 Mt Cu (Raesi et al. 2021).

The study area is the Shahr-e-Babak 1:100K geologic map (640 km²), which is located in South Eastern Iran (Fig. 1). The western parts of the area are covered with small outcrops of colored mélange units. A 150 m thick Eocene conglomerate unit along with a well-bedded and fine-grained sandstone unit and massive limestone units cover the eastern parts of the area. The Eocene flysch and volcanic units cover the study area in the north. The volcanic complex consists of andesitic basalt with high pyroxene and low olivine contents, as well as a 60 m thick red tuff, and trachy-andesite and trachy-basalt in lower parts.

The Eocene quartz monzonite and granodiorite rocks are outcropped in the South eastern. These rocks are similar in composition but quartz monzonite has more k-feldspar than plagioclase. The younger andesitic agglomerates cover volcanic rocks. The Neogene volcanic units outcrop around the Kuh-e-Masahim volcano in the north. This volcano forms the highest altitudes of the study area, with its peak having an elevation of 3500 m. An altered micro dioritic unit, which is the youngest subvolcanic unit in the area, forms the northern part of caldera of this volcano. The mafic units of the complex are also altered and contain sulfide mineralization. Figure 2 shows geology map of Shahr-e-Babak study area.

Dataset

In this study, we have used nine datasets to build the evidential maps: Geology, Geophysics, Geochemistry (Cu and PC 3), phyllic alteration, Argillic alteration, iron oxide alteration, lineaments density, and elevation data.

Geology

The 1:250K Shar-e-Babak geologic map was used as the basis for the lithological layer. This geologic map was digitized using the GIS software, and its Eocene and Neogene subvolcanic stocks, including diorite and granodiorite, were extracted as the heat sources (Fig. 3a).

Lineaments were extracted from the Geologic map and the ASTER dataset. Lineament density map was prepared from lineament structures (Fig. 3b). The density of lineaments in the north is more than other parts of the study area. The density map of these structures was prepared.

Geophysics

The airborne magnetic data acquired by the Atomic Energy Organization of Iran (AEOI) during 1977 and 1978 was used to prepare the total magnetic intensity map of the study area. These data were obtained at a flight spacing of about 500 m and an altitude of about 120 m. The RTP filter was performed on the total magnetic intensity map, and the anomalies were extracted from the resulted map (Fig. 3c).

Geochemistry

The data of stream sediment geochemical explorations can be used in Cu porphyry mineral potential modelling. A dataset of 604 samples from the geochemical surveys conducted by the GSI was used for geochemical processing. These samples were analyzed for Cu, Pb, Zn, Sb. Ni, Co, Cr, B and B. The boxplot method was applied to replace outlier data and the logarithmic method to normalize the remaining dataset. The histogram and QQ diagrams were plotted for the normalized data. The Cu anomaly map was created as an evidential layer for porphyry copper deposit prospectivity (Fig. 3d). The high Cu anomaly signatures are located in the north western and eastern parts of the study area. Also, a multivariate geochemical analysis (Principal Component analysis – PCA) was carried out on geochemical data. After preprocessing the geochemical data, PCA was applied on normalized data. PC3 that shows higher values of Cu, Pb, and Zn was determined to be associated with porphyry mineralization (Table 1). The map for this factor was also created as another evidential layer. Multivariate geochemical map is shown in Fig. 3e.

Table 1

Principal component analysis for multi-element geochemical data, where component 3 corresponds to Cu mineralization.
	Component
	1	2	3
Ni	.883	− .047	.136
Ba	− .016	.818	− .056
Pb	.153	.468	.695
B	.714	− .080	.258
Co	.842	− .065	− .080
Sb	− .110	.685	.098
Cr	.615	.557	.157
Cu	.521	.295	.515
Zn	.059	− .150	.808

ASTER data

The most significant advantages of using remote sensing in porphyry Cu mineralization exploration studies is the close relationship between alteration and mineralization areas (Riahi et al. 2022). Phyllic, argillic, and iron oxide alterations are common zones in cu porphyry systems. In this study, the mentioned alterations were obtained using the band ratio method from ASTER data. Band ratio is a widely used approach in improving the spectral characteristics of the alteration zones depending on the absorption bands of their altered minerals (Elhusseiny 2023).

The ASTER dataset covers a wide spectral range of 14 spectral bands, measuring reflected radiation in three bands between 0.52 and 0.86 µm (visible-near infrared, VNIR) with 15-m resolution, and six bands from 1.6 to 2.43 µm (shortwave Infrared, SWIR) with 30-m resolution. The emitted radiation is measured at 90-m resolution in five bands through the 8.125–11.65-µm wavelength region (thermal infrared; TIR) (Zoheir et al. 2019).

Band ratio 3/1 was applied for detecting the iron oxide alteration zone in the study area. Band ratios 7/6 and (5 + 7)/6 were applied for extracting argillic and phyllic alteration zones, respectively (Fig. 3f, 3g, 3h).

Elevation data

Volcanic rocks host the porphyry mineralization in the Shahr-e-Babak study area. They are used as dimension stones and have a rough topography. Therefore, elevation data were used as an effective evidential layer in modelling the host rocks of the porphyry mineralization in this area. The elevation data were extracted from ASTER images to prepare the Digital Elevation Model (DEM) (Fig. 3i).

A raster file with a cell size of 100×100 was generated for each evidential layer. Every evidential layer was then transferred into a [0, 1] interval using a linear fuzzy function. All of these raster files were transformed into an ASCII format to be used in MATLAB.

Target variable

In order to create target variables in the study area, we used 37 known mineralized points. Copper mines and indications were simply considered as known mineralized points, and a value of one was attributed to these points. On the other hand, non- deposit points refer to the points where there is no mineralization, and it is very complicated to specify whether a point is non-depositor not. Three conditions should be applied to create non-deposit points (Ghezelbash et al. 2021).

(1) The number of non-deposit and deposit points should be equal, which is 37 in this study.

(2) Non-deposit and deposit points should have an appropriate distance from each other.

(3) The deposit points usually occur in clusters or follow a specific form. Non-deposit points should be far from these clusters and should be chosen randomly.

The point pattern analysis was applied to delineate adequate distances between deposit and non-deposit points. Based on the point pattern analysis (Fig. 3j), all of the known mineral occurrences are located in a radius of 5247 m. As a result, with 100% probability, another occurrence can situate in this distance. Therefore, this distance was applied to known mineralized points to create the buffer distance. Also, in order to add accuracy to non-deposit points, a buffer of 500 m was applied to subvolcanic units as the heat sources and volcanic units as the host rocks of porphyry mineralization in the study area.

Artificial Neural Network

The ANNs methods have proved to be effective techniques in solving MPM problems with large datasets. The ANN method was first developed by computer scientists based on biological manner of neurons in the human brain. This method was later used in geoscience studies to help geoscientists to solve MPM problems.

There are different types of ANNs. In this paper, the multilayer perceptron (MLP) neural network has been used as a commonly used approach in geosciences. The structure of multilayer perceptron is shown in Fig. 4. MLP is a feed forward neural network (FFNN). The task of each neuron in the multilayer perceptron is to receive input data, process them, and send the processed data as output signal. At least, each neuron is connected with another neuron. Each MLP contains at least one hidden layer. One of the most important applications of MLP is solving highly nonlinear problems (Wang et al. 2020).

Adaptive Neuro fuzzy

Adaptive Neuro fuzzy inference system (ANFIS) was named from a combination of ANN and fuzzy systems. Actually, the ANFIS method benefits from the advantages of both ANN and fuzzy systems. In other words, in the ANFIS structure, the ANN method is used to improve the application of the fuzzy method. By using IF-THEN rules, the ANFIS method creates reasonable predictions from different complicated problems (Mehrabi et al. 2020). Two algorithms were applied to optimize the ANFIS results, including back propagation and Hybrid.

The ANFIS method can be summarized in five steps. First, the values of the membership function are calculated for input data. In other words, input data in the first layer are fuzzified. Second, the rule firing strength is calculated, and then, the input data are normalized. In the fourth step, the initial results of each fuzzy rule are generated, and finally the outputs of ANFIS are produced in the fifth layer. The structure of adaptive neuro fuzzy is shown in Fig. 5.

Random Forest

The random forest (RF) algorithm is a collaborative learning that was introduced by computer scientists based on Decision Trees (Taha et al. 2023). The algorithm produces many trees to optimize the model's performance (Keykhah et al. 2020). An RF classifier consists of some trees that are grown non-randomized. This growth process is based on the training dataset. For this process, a procedure called bootstrap bagging is applied, in which two-thirds of the data are used for training the algorithm, and the remaining are used to validate the accuracy of the model.

For prohibiting the over-fitting in the the training stage, RF produces trees with appropriate variables in the bag samples. This procedure decreases the correlation between trees and minimizes error in the training stages (Taha et al. 2023).

The structure of random forest is shown in Fig. 6.

General Regression neural network

The General neural network (GRNN) is a radial basis function neural network (RBFNN). GRNN is a highly RBFNN that operates based on one-pass algorithm and can produce continuous outputs (Martinez et al. 2022). GRNN can approximate implied results based on training data. Even if the input data is limited, GRNN can approximate optimal regressions (Liang et al. 2019). GRNN has been used in many prediction fields. In this method, the risk of trapping in the local minimum is decreased, and the capability of generalization as well as the learning rate is strengthened. GRNN contains four layers: input, pattern, summation, and output (Salgado et al. 2020). The task of the input layer is to receive data, transform it, and then send it to the pattern part. The neuron values are equal to input data. In the next step, the Euclidean distance between each input data and the related neuron is calculated. Therefore, there is a special neuron for input data. The calculated Euclidean distances are considered as the activation function. Next, these distances are sent to the next layer, which includes two neurons. The pattern layer weights are calculated in the summation part, and finally, resulted values are transferred to the output layer. The structure of GRNN is shown in Fig. 7.

This paper examines the performance of four machine learning methods (multi- layer perceptron, generalized regression neural network, random forest, and adaptive neuro fuzzy inference system) for producing mineral prospectivity map. In order to achieve this aim, the Shahr-e-Babak area was selected, which is a part of porphyry copper mineralization Uremia-Dokhtar belt. A total of 37 porphyry copper occurrence points were selected for training these models. Also, the point pattern analysis was used for delineating non-deposit points. Based on this method, 37 non-deposit points were selected, to be used in training the four models of machine learning methods.

Statistical correlation

The statistical correlation between predictive maps is presented in Fig. 8. The highest correlation is between Cu geochemical anomaly map and the PC3 geochemical anomalies. The correlations between Iron oxide, phyllic, and argillic alterations are high. The correlation between argillic - phyllic alterations, argillic -iron oxide alterations, and iron oxide - phyllic alterations are 0.69, 0.52, and 0.63, respectively. Also, the correlation between lineaments structures and argillic, phyllic, and iron oxide alterations are high. The correlation between argillic alteration-faults, phyllic alteration- faults, and iron oxide alteration- faults are 0.4, 0.66, and 0.6, respectively. This even shows that the faults have had a crucial role in transporting mineralizing solutions to the surface and producing the porphyry copper mineralization in the study area. The correlation between the RTP evidential layer and other evidential layers is negative. The correlation between RTP and phyllic, gossan, fault, factor 1, Cu, and argillic alteration layers are − 0.27, -0.37, -0.027, -0.09, -0.02, and − 0.021, respectively. These correlation coefficients show that the alterations caused demagnetized in the host rocks in the study area. The positive correlation between the alteration evidential layers and the Cu and PC3 geochemical anomalies shows the effective role of these zones in delineating Cu mineralization.

Spatial correlation

Figure 9 and Table 2 illustrate the spatial correlation between the evidential layers and the known Cu occurrences. Analyzing these plots shows significant correlation between known Cu mineralization systems and Cu geochemical anomalies, lineament structures, the RTP map, and phyllic, argillic, and iron oxide alterations, as well as factor 1 geochemical anomalies. The highest spatial correlation belongs to Cu anomalies and fault zones (0.71 and 0.49, respectively). As a result, the fault zone has played essential role in mineralization in the study area. Also, this confirms the result of statistical correlations presented in Fig. 8. The next highest spatial correlation belongs to the phyllic alteration (0.62) because this zone is formed near the center of porphyry deposits. With a value of 0.4, the spatial correlation of argillic alteration zones ranks next. The spatial correlation between iron oxide alteration and known mineralization points was equal to 0.32. The spatial correlations between known mineralization points and Elevation Model and aeromagnetic anomalies were 0.32 and 0.2. The low spatial correlation between aeromagnetic anomalies and mineralization points shows the extent of alteration and demagnetization in the study area. Based on alteration evidential layers, phyllic, argillic, and iron oxide zones are extended in the study area. These alterations caused demagnetization in volcanic host rocks in the study area. Therefore, the spatial correlation between aeromagnetic anomalies and known occurrences is equal to 0.2. All of these values show the positive role of evidential layers in porphyry copper prospectivity modelling.

Table 2

Input weight extracted from P-A plots
	Weight	Mean	Standard deviation
Cu	0.71	0.42	0.17
Phyllic	0.62
Lineaments	0.49
Intrusive bodies	0.49
Argillic	0.4
Iron oxide	0.32
DEM	0.32
Factor 1	0.25
RTP	0.2

Creating models with machine learning methods

As earlier mentioned, four machine learning methods, including MLP, ANFIS, RF, and GRNN were applied for detecting high potential areas.

The dataset was divided into training and testing data. The network was trained through different moods and ratios of training and testing data. Nine input layers were applied in the statistical and spatial correlations, including intrusive bodies, geochemical signatures (Cu and factor 3), magnetic anomalies, three alteration types (phyllic, argillic, and iron oxide), lineament structures, and the digital elevation model. All of the input data have positive spatial correlations with known Cu porphyry occurrences. Therefore, these evidential layers have positive input weights in training the model. Creating the models with machine learning methods are shown in Fig. 10. All of these models are compatible with each other. In all of the models, the high potential areas are located in western and eastern parts of the study area.

Model evaluation

In order to compare and classify the prospectivity models of the four machine learning methods, we applied natural break. The P-A method was also used for determining the optimal operation among the four methods. As seen in Fig. 11 and Table 3, the P-A plot values for MLP, ANFIS, RF, and GRNN are 0.85, 0.78, 1.27, and 0.76, respectively. The highest intersection happened for random forest algorithm and all weights of the models are higher than that of evidential layers.

This shows that the high favorability anomaly in the RF is minimum when the number of known occurrences is maximum.

For further evaluation, the fractal method was applied (Fig. 12), based on which the threshold values of the four machine learning methods were determined, and every potential model was subdivided into low potential, moderate potential, and high potential areas. Moderate and high values devoted to favorability prospective areas. The resulted maps based on fractal method are shown in Fig. 13.

A statistical evaluation of every threshold extent is presented in Fig. 14. Based on this diagram, the highest favorability area belongs to the random forest algorithm that is equal to 16%. The high favorability of multilayer perceptron model occupies 21.6% of the study area. This is slightly lower than that of the random forest method. The high favorability areas of ANFIS and GRNN algorithms occupy 35% and 35.2% of the study area, which are nearly equal and have a simultaneous operation.

In the final step, the models resulted from the four machine learning methods were integrated, and using the fractal method, this model was divided into background, moderate potential areas, and high potential areas (Fig. 15). The favorability area contains known Cu occurrences that are extended along linear structure zones. This coincidence implies the role of faults and fractured zones in Cu mineralization in the Shahr-e-Babak study area.

Based on these four machine learning methods, the proper favorability areas are located in the eastern and western parts of the study area. There is a high potential target in the north that is located in the flysch host rocks. This target coincides with high magnetic anomalies in sedimentary rocks, which means the anomaly belongs to an igneous rocks. This target has not been properly explored, so it can be introduced as an appropriate target in the Shahr-e-Babak study area.

Table 3

Weight extracted from P-A plots
	weight	mean	Standard deviation
Random Forest	1.26	0.89	0.23
Generalized Neural Network	0.76
Multi Layer Percepteron	0.85
ANFIS	0.78

Mineral Prospectivity Modelling (MPM) is a useful tool for highlighting areas of interest in challenging exploration settings. This modelling process was conducted on the Shahr-e-Babak area (Southern Iran), using a combination of nine layers of predictive maps. Four machine learning algorithms, including multilayer perceptron, adaptive neuro fuzzy inference system, random forest, and generalized neural network methods were used for porphyry copper modelling in the study area. There are good correlations between argillic, phyllic, and iron oxide alterations with structural lineaments. This confirms the role of faults and structures in the formation of alterations in the area. The low spatial correlation between RTP magnetic anomalies and Cu occurrences implies demagnetization of rocks through alteration processes. The weights of phyllic, argillic, and iron oxide alterations imply alteration zoning in the study area. The P-A plot method shows the positive role of evidential layers in Cu porphyry prospectivity modelling in this study. The resulted model of machine learning methods imply the high accuracy of these methods. The resulted weights of the P-A plot for multilayer perceptron, adaptive neuro fuzzy inference system, random forest and generalized neural network are 0.85, 0.78, 1.27, and 0.76, respectively. These results reflect the high accuracy of the prospectivity models. The main porphyry copper prospect was detected in the western and eastern parts of the study area.

Data availability

Not applicable.

Funding

The authors did not receive support from any organization for the submitted work.

Acknowledgements

The authors would like to thank M. Azadi, the Editor-in-chief, for handling this manuscript. We also thank the anonymous reviewer for his/her constructive comments.

Contributions

All authors contributed to the study conception and design. Data collection and modeling were performed by Moslem Jahantigh and Hamidreza Ramazi. The first draft of the manuscript was written by Moslem Jahantigh. Hamid Reza Ramazi read and approved the final manuscript.

Corresponding author

Correspondence to Hamidreza Ramazi.

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Daviran M, Maghsoudi A, Ghezelbash R, Pradhan B (2021) A new strategy for spatial predictive mapping of mineral prospectivity: Automated hyperparameter tuning of random forest approach. Comput Geosci v.148:104688
Daviran M, Parsa M, Maghsoudi A, Ghezelbash R (2022) Quantifying Uncertainties Linked to the Diversity of Mathematical Frameworks in Knowledge-Driven Mineral Prospectivity Mapping. Nat Resour Res 31, 2271–2287.
Daviran M, Shamekhi M, Ghezelbash R, Maghsoudi M (2022) Landslide susceptibility prediction using artificial neural networks, SVMs and random forest: hyperparameters tuning by genetic optimization algorithm. Int J Environ Sci Technol 1, 259-276.
Elhusseiny A (2023) Integrated Structure and Mineralization Study Using Aero-Magnetic, Aero-Spectrometric and Remote Sensing Data at Esh El-Mallaha Area. Eastern Desert, Egypt. Geomaterials, 13, 1-22.
Ghezelbash R, Maghsoudi A, Shamekhi M (2023). Genetic algorithm to optimize the SVM and K-means algorithms for mapping of mineral prospectivity. Neural Comput & Applic 35, 719–733.
Ghezelbash R, Maghsoudi A, Bigdeli, A (2021) Regional-Scale Mineral Prospectivity Mapping: Support Vector Machines and an Improved Data-Driven Multi-criteria Decision-Making Technique. Nat Resour Res 30, 1977–2005.
Keykhah M, Kohsary A, Morshedy M, Porwal A (2020) A machine learning-based approach to exploration targeting of porphyry Cu-Au deposits in the Dehsalm district, eastern Iran. Ore Geol. Rev, 116, 103234.
Liang Y, Niu D, Hong W (2019) Short term load forecasting based on feature extraction and improved general regression neural network model. Energy, 166,653-663.
Martínez F, Charte F, Frías M, María A, Rodríguez M (2022) Strategies for time series forecasting with generalized regression neural networks. Neurocomputing, 491, 509-521.
Mehrabi M, Pradhan B, Moayedi H, Alamri A (2020) Optimizing an Adaptive Neuro-Fuzzy Inference System for Spatial Prediction of Landslide Susceptibility Using Four State-of-the-art Metaheuristic Techniques. Sensors 20(6):17-23.
Raeisi D, Zhao M, Babazadeh S, Long L, Hajsadeghi S, Modabberi S (2021) Synthesis on productive, sub-productive and barren intrusions in the Urumieh-Dokhtar magmatic arc, Iran, constraints on geochronology and geochemistry, Ore Geol. Rev, 132.
Riahi S, Bahroudi A, Abedi M, Aslani S, Lentz D (2022) Evidential data integration to produce porphyry Cu prospectivity map, using a combination of knowledge and data‐driven methods. Geophys, 70 (2), 421 – 437.
Salgado C, Dam R, Salgado W, Werneck R, Pereira C, Schirru R (2020) The comparison of different multilayer perceptron and General Regression Neural Networks for volume fraction prediction using MCNPX code. Appl Radiat Isot, 162, 109- 170.
Sun T, Chen F, Zhong L, Liu W, Wang Y (2019) GIS-based mineral prospectivity mapping using machine learning methods: A case study from Tongling ore district, eastern China. Ore Geol. Rev, 109, 26-49.
Taha M, Xi Y, Hu A, Wang S, Liu X (2023) Investigating the Capabilities of Various Multispectral Remote Sensors Data to Map Mineral Prospectivity Based on Random Forest Predictive Model: A Case Study for Gold Deposits in Hamissana Area. NE Sudan. Minerals 13, 49.
Wang X, Lu H, Wei X, Wei G, Behbahani S, Iseley T (2020) Application of Artificial Neural Network in Tunnel Engineering: A Systematic Review. in IEEE Access. 8, 119527-119543.
Zoheir B, Emam A, Abdel-Wahed, Soliman N (2019) Multispectral and Radar Data for the Setting of Gold Mineralization in the South-Eastern Desert Egypt. Remote Sensing, 11,1450.

No competing interests reported.

Porphyry Copper Prospectivity using Machine Learning Methods: A Case Study of the Shahr-e-Babak Prospecting Area, South Eastern Iran

Status:

Version 1

Abstract

Figures

Introduction

Study area and data used

Geology

Dataset

Geology

Geophysics

Geochemistry

ASTER data

Elevation data

Target variable

Methods

Artificial Neural Network

Adaptive Neuro fuzzy

Random Forest

General Regression neural network

Results and discussion

Statistical correlation

Spatial correlation

Creating models with machine learning methods

Model evaluation

Conclusion

Declarations

References

Additional Declarations

Status:

Version 1