The Evolution of Machine Learning in Large-Scale Mineral Prospectivity Prediction: A Decade of Innovation (2016–2025)

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Dear authors, thanks for submitting your work to Minerals. 

There are several points which need clarification/improvement:

line 51 > "Traditional methods rely on experts' experience and linear regression models".
this is not true. there are geostatistical and simulation methods which are not based on linear regression. 

line 56 > "Moreover, traditional methods are inefficient in processing large datasets and lack objective quantification and uncertainty assessment of predictions, restricting their application in large-scale, high-throughput exploration".
this is indeed not true. there are simulation methods, such as SGS which provide the uncertainty for every prediction. 

Line 98 > Autoencoders are not new. The method introduced about forty years ago, but recently is using as a deep learning technique for mineral exploration/modelling. 

We need a space before reference numbers 1 to 12 in the main text.

Line 208: Fig.1 > the first row of the figure is irrelevant to the submitted manuscript as it is showing "Vibration and Acoustics"! with a total of 96 documents. 

Several references are not adequately described in the main text; e.g. references number 23 to 27. 

Line 318 > this figure needs improvements; e.g. naming machine learning methods for each mineral prediction scenario. 

Line 319> The caption of Fig. 2 is too short and non-informative. 

Please use the full name of ML methods OR their abbreviation, not a combination of the two.

Some references are not ordered truly in the main text; e.g. references number 38, 41 and 44. 

There are many abbreviations in the text, which are not presented by their full name. Examples are: SHAP, LINE and XGBoost. 

Line 707 > it is not time dependent! there are always a need for such reliable methods. the grow of the ML studies reflects the interest among researchers. 

There are several references in the conclusions. If we are concluding from our work, it should focus on the obtained results, not those from outside.
One more important note about the conclusions section: we need to complete the concluding part with summarized numerical results. It is mandatory to have the results in the conclusions. 

There are many page number missing and not using a unique style in the references. Some of them are marked by the red lines. Please unify the referencing style and complete the details.

For more details, please refer to the attached PDF file. 

Comments for author File: Comments.pdf

Author Response

Comment 1: Line 51 > "Traditional methods rely on experts' experience and linear regression models." this is not true. there are geostatistical and simulation methods which are not based on linear regression.

Response 1: Thank you for pointing this out. We agree with this comment. Therefore, we have revised the statement to acknowledge the existence of non-linear regression-based methods like geostatistics and simulation. The revision can be found on page 2, paragraph 2, lines 54-58: "Traditional methods, such as linear regression models, have difficulty capturing complex nonlinear relationships. However, there are other methods, such as geostatistical methods (for example, the Kriging method) and simulation methods (for example, Monte Carlo simulation), which do not rely on linear regression but still have limitations when dealing with complex geological data ."

 

Comment 2: Line 56 > "Moreover, traditional methods are inefficient in processing large datasets and lack objective quantification and uncertainty assessment of predictions, restricting their application in large-scale, high-throughput exploration." this is indeed not true. there are simulation methods, such as SGS which provide the uncertainty for every prediction. Response 2: We appreciate your insightful feedback. We concur that our original statement was misleading. We have modified the text to clarify that while traditional methods may face challenges in processing large datasets efficiently, simulation methods like SGS do offer uncertainty assessments. The updated text is located on page 2, paragraph 2, lines 61-66: "Moreover, traditional methods are inefficient in processing large datasets and lack objective quantification and uncertainty assessment of predictions, restricting their application in large-scale, high-throughput exploration [11]. While some traditional methods, such as Sequential Gaussian Simulation (SGS), can provide uncertainty assessments for predictions, they often struggle with the complexity and scale of modern geological datasets."

 

Comment 3: Line 98 > Autoencoders are not new. The method introduced about forty years ago, but recently is using as a deep learning technique for mineral exploration/modelling.

Response 3: Thank you for highlighting this clarification. We acknowledge the oversight and have revised the text to accurately reflect the history of autoencoders. The correction is on page 3, paragraph 5, lines 106-109: "It is worth noting that while autoencoders have been recently applied as a deep learning technique in mineral exploration and modelling, they are not new—this method was introduced approximately forty years ago."

 

Comment 4: We need a space before reference numbers 1 to 12 in the main text.

Response 4: We appreciate your attention to detail. We have carefully reviewed the main text and added the necessary space before each of the reference numbers 1 to 12. These corrections can be found throughout the manuscript, specifically in lines 58 63 68 etc.

 

Comment 5: Line 208: Fig.1 > the first row of the figure is irrelevant to the submitted manuscript as it is showing "Vibration and Acoustics"! with a total of 96 documents.

Response 5: Thank you for bringing this to our attention. We have inspected Figure 1 and removed the irrelevant first row related to "Vibration and Acoustics." The revised Figure 1 now accurately reflects the content of the submitted manuscript and can be viewed on page 6, Figure 1.

 

Comment 6: Several references are not adequately described in the main text; e.g. references number 23 to 27.

Response 6: We value your feedback and have thoroughly reviewed references 23 to 27. We have enhanced the descriptions of these references in the main text to provide better context and relevance. The improvements can be found in lines 250-265 ., where each reference is discussed in more detail.

 

Comment 7: Line 318 > this figure needs improvements; e.g. naming machine learning methods for each mineral prediction scenario.

Response 7: We are grateful for your suggestion. We have revised Figure 2 by clearly naming the machine learning methods associated with each mineral prediction scenario. The updated figure is presented on page 8,in lines 347, Figure 2, ensuring that readers can easily identify which methods correspond to specific scenarios.

 

Comment 8: Line 319> The caption of Fig. 2 is too short and non-informative. Please use the full name of ML methods OR their abbreviation, not a combination of the two.

Response 8: Thank you for this valuable input. We have expanded the caption of Figure 2 to be more informative and have standardized the naming convention for machine learning methods, using their full names for clarity. The revised caption can be found on page 8, in lines 349-357: "This network diagram, with machine learning techniques at its core and various application scenarios distributed peripherally, illustrates the connections between typical algorithms—such as CNN, SVM, and RNN—and application contexts like mineralization anomaly identification and target area selection."

 

 

Comment 9: Some references are not ordered truly in the main text; e.g. references number 38, 41 and 44.

Response 9: We appreciate your observation. We have meticulously checked the order of references 38, 41, and 44 in the main text and have corrected any discrepancies. The references now appear in the correct order in lines 471-485.

 

Comment 10: There are many abbreviations in the text, which are not presented by their full name. Examples are: SHAP, LINE and XGBoost.

Response 10: We thank you for pointing this out. We have gone through the manuscript and ensured that all abbreviations—such as SHapley Additive exPlanations (SHAP), Large-scale Information Network Embedding (LINE), and eXtreme Gradient Boosting (XGBoost)—are first introduced with their full names followed by the abbreviation in parentheses.This change can be found in lines 519 586, where the abbreviations are first mentioned.:"like SHapley Additive exPlanations(SHAP) , were widely used in deep learning models.eXtreme Gradient Boosting(XGBoost)"

 

Comment 11: Line 707 > it is not time dependent! there are always a need for such reliable methods. the grow of the ML studies reflects the interest among researchers.

Response 11: We appreciate your clarification. We have rephrased the statement on line 707 to remove the implication of time dependence and to better reflect the ongoing need for reliable methods. The revised text is on page 18, paragraph 5, lines 748-754: "This growth reflects the global surge of interest in artificial intelligence and the expanding availability of high-quality geoscience data. It mirrors the increasing enthusiasm and capacity of the research community to adopt and test machine-learning approaches. The continuous increase in literature is not just a simple accumulation of research output but also represents an expansion in research depth and breadth, indicating that machine learning will play a more central role in future mineral prediction."

 

Comment 12: There are several references in the conclusions. If we are concluding from our work, it should focus on the obtained results, not those from outside.

Response 12: We are thankful for your guidance. We have restructured the conclusions section to focus specifically on the results obtained from our study, minimizing references to external work. The updated conclusions can be found on page 30-32, where we summarize our key findings without emphasizing external contributions.

 

Comment 13: One more important note about the conclusions section: we need to complete the concluding part with summarized numerical results. It is mandatory to have the results in the conclusions.

Response 13: We understand the importance of including numerical results in the conclusions and have made the necessary additions. The concluding section now incorporates key numerical outcomes from our research on page 30-32, ensuring that the conclusions are data-driven and reflective of our study's contributions.

 

Comment 14: There are many page number missing and not using a unique style in the references. Some of them are marked by the red lines. Please unify the referencing style and complete the details. For more details, please refer to the attached PDF file.

Response 14:Thank you for your detailed feedback. We have carefully reviewed the references section as suggested. Specifically, we have supplemented the page numbers for some entries where this information was previously missing. However, for a portion of the references—such as certain electronic resources and open-access articles—the original sources themselves do not provide complete page number details (these types of materials often omit page numbering by nature). That said, we have verified through checks that all these references are fully accessible and retrievable. To the best of our ability, we have completed the supplementation of missing page numbers.

Additionally, we have standardized the formatting of all references to ensure consistency in style, addressing the issue of inconsistent referencing styles you noted. We hope these revisions meet the required standards, and please feel free to let us know if further adjustments are needed.

 

Thank you very much for giving us the valuable opportunity to revise and improve our thesis. In response to all the opinions you raised, we have carefully revised and adjusted each one to ensure that every issue is properly resolved. For more details about this revision, such as specific content adjustments and format optimizations, please refer to the PDF file in the attachment. Once again, we sincerely thank you for the time and effort you have devoted to the review of the thesis. Your professional feedback not only helped us correct the existing problems, but also has significant guiding significance for improving the overall quality, rigor and accuracy of the thesis. We are looking forward to receiving your further feedback so that we can supplement and improve it when necessary to push the paper to a higher level.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This manuscript reviews the evolution of machine learning applications in large-scale mineral prospectivity prediction from 2016–2025. The topic is timely, relevant, and of significant value to both the academic and professional exploration community. The paper demonstrates strong effort in literature collection (255 publications) and provides a comprehensive bibliometric overview of methods, applications, and trends.

That said, I see several areas where the manuscript could be strengthened before publication:

1. Critical depth: The current version is largely descriptive and bibliometric. Please expand the analysis to critically assess methodological soundness, reproducibility, and real-world applicability. For instance, claims about performance improvements (e.g., CNNs >95% accuracy, hybrids improving 15–20%) should be supported with explicit case details or quantitative evidence to support the statements. 

2. Geographical scope: Restricting analysis to nine countries (China, Canada, Australia, USA, Iran, South Africa, Brazil, Finland, India) risks bias. Important contributions from Latin America and Europe are overlooked. At minimum, this limitation should be acknowledged explicitly, and additional international works should be cited.

3. References and inclusivity: Expand citations to include foundational and critical works beyond the selected countries. This will reduce regional bias and increase the authority of the review.

4. Clarity and conciseness: The English is understandable but often verbose and repetitive. A language edit focusing on conciseness would improve readability.

5. Balance Between ML and Domain Knowledge: The narrative sometimes suggests that ML alone can solve mineral prediction challenges. In reality, success depends on geological knowledge integration (e.g., structural geology, alteration models).The review should more critically discuss how domain expertise constrains and validates ML outputs, avoiding an overly technology-driven perspective.

Overall, this paper has the potential to become a valuable reference, but it requires revisions to deepen its critical perspective, broaden its inclusivity, and strengthen its evidence base. 

Author Response

Comments 1: The current version is largely descriptive and bibliometric. Please expand the analysis to critically assess methodological soundness, reproducibility, and real-world applicability. For instance, claims about performance improvements (e.g., CNNs >95% accuracy, hybrids improving 15–20%) should be supported with explicit case details or quantitative evidence to support the statements.

Response 1: We appreciate the suggestion to strengthen the critical analysis of methodological soundness and real-world applicability. To address this, we have expanded the discussion in Sections 3.1 and 3.2 to include explicit case details and quantitative evidence for claims regarding performance improvements. For example, we now provide specific case studies where CNNs achieved >95% accuracy in remote sensing alteration mineral mapping, supported by quantitative metrics such as precision, recall, and F1-scores (see revised page 14, paragraph 1). Additionally, we have added details on the reproducibility of hybrid models (e.g., CNN-SVM), including improvements of 15–20% in accuracy over single-method approaches, with explicit references to studies demonstrating these results (see revised page 14, paragraph 3).

 

Comments 2: Restricting analysis to nine countries (China, Canada, Australia, USA, Iran, South Africa, Brazil, Finland, India) risks bias. Important contributions from Latin America and Europe are overlooked. At minimum, this limitation should be acknowledged explicitly, and additional international works should be cited.

Response 2: We acknowledge the limitation of geographical bias in our analysis and appreciate the suggestion to broaden the scope. We have explicitly acknowledged this limitation in Section 2.1 (see revised page 5, paragraphs 3-5) and have expanded our citations to include significant contributions from Latin America and Europe. For example, we now reference studies from countries such as Germany, France, the United Kingdom, Chile and Peru, which highlight the application of machine learning in mineral prediction in these regions (see revised pages 22-24, and updated references 45, 56, 57-63and 67).

 

Comments 3: Expand citations to include foundational and critical works beyond the selected countries. This will reduce regional bias and increase the authority of the review.

Response 3: We agree that expanding the citations to include foundational and critical works from a broader range of countries enhances the review’s authority. We have added key references from Europe and Latin America, such as seminal works on geostatistics and simulation methods that predate machine learning applications in mineral prediction. We have also revised the reference list to ensure inclusivity and reduce regional bias.(The main update contents see revised pages 22-24, and updated references 45, 56, 57-63and 67).

 

Comments 4: The English is understandable but often verbose and repetitive. A language edit focusing on conciseness would improve readability.

Response 4: We appreciate the feedback on the language and have performed a comprehensive edit to improve conciseness and readability. Redundant phrases and repetitive statements have been streamlined throughout the manuscript. For example, in Section 3.4, we condensed the discussion on the application fields of machine learning methods while retaining all critical information (see revised pages 22–23). We believe these revisions significantly enhance the clarity of our work.

 

Comments 5: The narrative sometimes suggests that ML alone can solve mineral prediction challenges. In reality, success depends on geological knowledge integration (e.g., structural geology, alteration models). The review should more critically discuss how domain expertise constrains and validates ML outputs, avoiding an overly technology-driven perspective.

Response 5: We agree that the integration of geological knowledge is critical to the success of machine learning in mineral prediction. To address this, we have revised Sections 3.2 and 4.1 to emphasize the role of domain expertise in constraining and validating ML outputs. We now include specific examples of how geological insights (e.g., structural geology, alteration modeling) are combined with ML techniques to improve prediction accuracy and reliability (see revised pages 14–15 and 25–26). We have also adjusted the narrative to avoid an overly technology-driven perspective, ensuring a balanced discussion of ML and domain knowledge.

 

We sincerely appreciate you giving us the valuable opportunity to revise and refine our paper. In response to all the comments you put forward, this revision aims to enhance the paper's critical depth, inclusiveness, and evidential basis. For detailed information regarding the specific revisions and format optimizations, please refer to the PDF file in the attachment. We would like to thank you again for the time and effort you have devoted to reviewing the paper, and we look forward to receiving your further feedback. This will enable us to make supplementary improvements if necessary and elevate the paper to a higher standard.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

I read the paper with great interest, and I appreciate the effort to provide a comprehensive review of machine learning applications in geosciences over the past decade. However, I believe that the scope and methodology of the study could be significantly improved to better reflect the global landscape and recent developments.

First, the geographical coverage of the statistics appears unbalanced. The analysis highlights contributions from countries such as China, India, the USA, Australia, Canada, and other countries, with only Finland representing Europe. This exclusion overlooks the substantial and growing body of research originating from many other European countries and Russia, where innovative applications of machine learning in geosciences and mining industry have been steadily advancing. For a review that aims to offer a worldwide perspective, such omissions risk biasing the conclusions and underestimating important regional contributions.

Second, while the paper provides a useful overview of traditional and well-established machine learning methods (e.g., Random Forests, Decision Trees, Autoencoders etc., and Deep Neural Networks), it does not address the transformative impact of Large Language Models (LLMs), which have emerged as a major force in geoscientific research in the last few years. Recent literature demonstrates how LLMs are being applied to document mining, automated knowledge extraction, multimodal integration, and geoscientific mapping. These developments are reshaping the methodological landscape of the field and deserve to be recognized in any up-to-date review.

In addition, the Authors just mention Reinforcement Learning techniques, without including effectively that family of Machine Learning techniques in their analysis.

In summary, while the paper makes a valuable contribution, its reliance on a geographically narrow dataset and its omission of LLM-based approaches, as well as Reinforcement Learning methods (just mentioned few times), limit the representativeness of its conclusions. A broader statistical base and an inclusion of recent paradigm-shifting technologies would provide a more balanced and truly global assessment of machine learning in geosciences over the last decade.

Minor comment: please double check Table 2 (“Literature analysis” is repeated twice).

Author Response

Comment 1:The geographical coverage of the statistics appears unbalanced. The analysis highlights contributions from countries such as China, India, the USA, Australia, Canada, and other countries, with only Finland representing Europe. This exclusion overlooks the substantial and growing body of research originating from many other European countries and Russia, where innovative applications of machine learning in geosciences and mining industry have been steadily advancing. For a review that aims to offer a worldwide perspective, such omissions risk biasing the conclusions and underestimating important regional contributions.

Response 1:Thank you for this insightful comment. We agree that the geographical coverage in the original manuscript was unbalanced and overlooked contributions from European countries and Russia. To address this, we have expanded the analysis to include research contributions from additional European countries (e.g., Germany, France, the UK) and Russia. We have updated Figure 6 and Table 3 to reflect a more comprehensive global landscape. The revised text now explicitly acknowledges the growing body of work from these regions and discusses their specific contributions to machine learning applications in geosciences. (see revised pages 22-24, and updated references 45, 56, 57-63and 67).

 

Comment 2:While the paper provides a useful overview of traditional and well-established machine learning methods (e.g., Random Forests, Decision Trees, Autoencoders, etc., and Deep Neural Networks), it does not address the transformative impact of Large Language Models (LLMs), which have emerged as a major force in geoscientific research in the last few years. Recent literature demonstrates how LLMs are being applied to document mining, automated knowledge extraction, multimodal integration, and geoscientific mapping. These developments are reshaping the methodological landscape of the field and deserve to be recognized in any up-to-date review.

Response 2:Thank you for highlighting this important gap. We acknowledge that Large Language Models (LLMs) have gained significant traction in geoscientific research recently. To address this omission, we added a new paragraph in Section 3.4, which is about "The Emerging Role of Large Language Models (LLMS)". This subsection discusses the applications of LLMs in document mining, knowledge extraction, and geoscientific mapping, supported by recent literature.These changes can be found on page 16, and lines 655-670.

 

Comment 3:The Authors just mention Reinforcement Learning techniques, without including effectively that family of Machine Learning techniques in their analysis.

Response 3:Thank you for this observation. We agree that Reinforcement Learning (RL) deserves greater attention in the context of machine learning applications in geosciences. To address this, we have expanded the discussion of RL in Section 3.4, emphasizing its potential in dynamic geological modeling and real-time exploration decision-making. These changes can be found on page 16, and lines 655-670.

It is worth noting that emerging paradigms such as Transformer models, Graph Neural Networks (GNNs), and self-supervised learning have also attracted widespread attention in the academic community. Through innovative modeling methods, these technologies can effectively handle the structural characteristics of time-series data, spatial data, and unlabeled data, providing important support for the construction of future mineral exploration and prediction systems. However, due to the fact that this field is still in a state of rapid development and constraints such as the timeline of this paper’s writing, this paper mainly focuses on relatively mature deep learning methods, resulting in fewer mentions of the aforementioned emerging paradigms. Nevertheless, the potential of RL in dynamic geological modeling and real-time exploration decision-making remains undeniable, and we have already stated this in the limitations section of the paper.

 

Comment 4:In summary, while the paper makes a valuable contribution, its reliance on a geographically narrow dataset and its omission of LLM-based approaches, as well as Reinforcement Learning methods (just mentioned few times), limit the representativeness of its conclusions. A broader statistical base and an inclusion of recent paradigm-shifting technologies would provide a more balanced and truly global assessment of machine learning in geosciences over the last decade.

Response 4:We highly value this comprehensive feedback. To address it, we have revised the methodology chapter to explicitly state that the current analysis has integrated a broader dataset, covering research findings from multiple European countries and Russia. Meanwhile, we have systematically organized the discussions related to Large Language Models (LLMs) and Reinforcement Learning (RL) throughout the entire paper, and have also made statements regarding the limitations of the paper and the aspects that were not covered. Our aim is to construct a more comprehensive and balanced evaluation framework. These improvements are reflected in the revised Section 2, Section 3.4, and lines 1266–1277, with Figure 7 and Table 3 also updated accordingly.

 

Minor Comment:Please double-check Table 2 (“Literature analysis” is repeated twice).

Response to Minor Comment:Thank you for pointing this out. We have reviewed Table 2 and corrected the repetition of “Literature analysis.” The table now accurately reflects the distribution of research methods. This change can be found on page 17, Table 2.

 

We sincerely appreciate you giving us the valuable opportunity to revise and refine our paper. In response to all the comments you put forward, this revision aims to enhance the paper's critical depth, inclusiveness, and evidential basis. For detailed information regarding the specific revisions and format optimizations, please refer to the PDF file in the attachment. We would like to thank you again for the time and effort you have devoted to reviewing the paper, and we look forward to receiving your further feedback. This will enable us to make supplementary improvements if necessary and elevate the paper to a higher standard.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

I appreciate the effort the authors have made in revising the manuscript. The major concerns raised in the first round have been carefully addressed, and the paper is now clearer, more comprehensive, and well balanced. The improvements in critical analysis, references, geographical scope, and integration with domain knowledge make this a solid and timely contribution.

I consider the revised version acceptable for publication.

Author Response

We would like to extend our sincere and heartfelt gratitude to you for your thorough review and positive feedback on the revised version of our manuscript. Your recognition of the efforts we dedicated to addressing the major concerns raised in the first round—along with your affirmation of the improvements in critical analysis, reference quality, geographical scope, and integration with domain knowledge—means a great deal to our team.