Identifying Conditioning Factors and Predictors of Conflict Likelihood for Machine Learning Models: A Literature Review

Round 1

Reviewer 1 Report

Introduction: I think it would be better to move the description of ACLED and UCDP datasets in the Introduction section to the Methodology section. In addition, you should highlight the significance of using machine learning techniques. The reason why you have to focus on machine learning models to predict the likelihood of conflicts is not presented in the Introduction section. The increasing use of machine learning in the field may be related to the reason you missed. You should also strengthen the reason why identifying predictors to predict the likelihood of conflicts is important. You should highlight it through literature review.

 

The description of ACLED and UCDP datasets in Section 3.1 should be moved to the Methodology section.

 

Identifying predictors with the number of citations and times each condition factor was used is not sufficient to suggest important features that researchers should consider for their research. It does not necessarily mean that it indicates its importance, and it can be related to something else, like availability of data. Therefore, I recommend examining the importance of features in the previous studies you selected if it is available, and you should make overall importance rankings of the features. There must be some studies that analyzed the importance, and It would be meaningful you summarize it to overall relative importance of features. In that way, you can suggest which features are more important than others.

Moreover, you should show a theoretical model to explain the relationships between conditioning factors and conflicts. You should explain the theories behind the relationships based on the theoretical model, and you should regroup the conditioning factors based on the categories of the model.

In addition, are there any factors that many researchers missed due to some reason, like data availability, but are important? It would be great if you mention those factors in the Conclusion and discussion section.

Sufficient

Author Response

Dear Reviewer,

Thank you very much for your great review and feedback. Your comments and suggestions have been duly noted, and we appreciate the constructive criticism, which will undoubtedly help us in improving our research paper. Revisions are highlighted in yellow in the attached file. Please find our response in red below:

Introduction: I think it would be better to move the description of ACLED and UCDP datasets in the Introduction section to the Methodology section. In addition, you should highlight the significance of using machine learning techniques. The reason why you have to focus on machine learning models to predict the likelihood of conflicts is not presented in the Introduction section. The increasing use of machine learning in the field may be related to the reason you missed. You should also strengthen the reason why identifying predictors to predict the likelihood of conflicts is important. You should highlight it through literature review.

The description of ACLED and UCDP datasets in Section 3.1 should be moved to the Methodology section.

As per your suggestion, we agree that the detailed descriptions of the ACLED and UCDP datasets would fit more coherently in the Methodology section of the paper. This will not only improve the flow of the introduction but also provide a more in-depth context when discussing the methods used in the research. Please refer to the lines 233 to 274 highlighted in yellow in the revised manuscript.

Your point on highlighting the significance of using machine learning techniques is well-taken. We revised the introduction to clearly underline the importance and effectiveness of machine learning models in conflict prediction. In the revised version, we elaborated on how machine learning's ability to handle vast data sets, recognize intricate patterns, and make accurate predictions makes it a potent tool in conflict prediction. We also emphasized the growing prevalence of machine learning in this field and illustrated it with a reference to the ViEWS project of Uppsala University and the INFORM Project from the Joint Research Center of the European Commission. Please refer to the lines 59 to 71 in the revised manuscript.

Your comment on strengthening the reason for identifying predictors to predict conflicts is very valuable. We included in the introduction a paragraph with a more in-depth literature review highlighting the critical need to identify predictors. This will showcase the research's urgency and necessity and will underscore the profound implications of accurately predicting conflicts. Please refer to the line 72 to 85 and lines 275 to 315.

Identifying predictors with the number of citations and times each condition factor was used is not sufficient to suggest important features that researchers should consider for their research. It does not necessarily mean that it indicates its importance, and it can be related to something else, like availability of data. Therefore, I recommend examining the importance of features in the previous studies you selected if it is available, and you should make overall importance rankings of the features. There must be some studies that analyzed the importance, and It would be meaningful you summarize it to overall relative importance of features. In that way, you can suggest which features are more important than others. Moreover, you should show a theoretical model to explain the relationships between conditioning factors and conflicts. You should explain the theories behind the relationships based on the theoretical model, and you should regroup the conditioning factors based on the categories of the model.

Thank you for your insightful comments. Your points have points on the complex relationship between the problem of identifying the priority and the significance of each conditioning factor. This is a complex relationship, and while various organizations such as JRC and Uppsala have used conditioning factors on a global or regional level in their attempt to develop predictive models, other scholars such as Clin Flint consider that the conditioning factors inherent complexities and unique geographical, social, political, and economic contexts of specific areas where conflicts are taking place. From my own experience in Darfur, providing information management services to the NGO community, I can suggest that one of the main components that contributed to the conflict in Darfur was access to grazing land. Thus, such a condition factor plays a significant role in the onset of the conflict in Darfur, though it may not be the case in other geographical locations where grazing land does not present such a valuable commodity. That we believe that the general ranking of variables might not always provide an applicable solution for predicting conflicts, as the importance and composition of these features depend on the conflict type, location, and specific local circumstances. Therefore, we used the list of citations, and the frequency of each conditioning factor as a foundational basis. This method helps identify which factors have generally been applied in machine learning and conflict prediction research. We also encourage researchers to thoroughly review the local conditions specific to their research area and consider the quality and availability of data and local specificities. Please refer to lines 275 to 315.

In addition, are there any factors that many researchers missed due to some reason, like data availability, but are important? It would be great if you mention those factors in the Conclusion and discussion section.

Thank you for this suggestion. We included a section in the Conclusion and Discussion where we state that “conflict dynamics are constantly evolving, and new types and elements may emerge in yet-to-be-studied scenarios. Additionally, certain conditioning factors may have been overlooked due to limitations in data availability during research studies. Therefore, we perceive this research as a process of living nature, adaptable and open to future enrichment with additional conditioning factors as the field of conflict prediction studies continues. As the application of machine learning to conflict prediction gains momentum, we anticipate an increase in the identification and understanding of critical conflict conditioning factors, further enhancing the accuracy and precision of conflict prediction models.” Please refer to lines 807 to 815.

We hope these changes would help to address your suggestions, we appreciate your thoughtful feedback as it guides us toward improving the quality and relevance of our work. We are grateful for your invaluable feedback and look forward to further comments and suggestions.

Best regards,

Timur Obukhov and Maria Brovelli.

Author Response File: Author Response.docx

Reviewer 2 Report

In this paper, the use based on a literature review identifie the conditioning factors and predictors of conflict like-lihood for machine learning models. The work is well presented and will consititute a great reference for machine learning in conflict predictors in the future.

Two small comments:

From the title is not evident that the article is based on a literature review and no experiment is performed.

Authors should provide the list of papers utlizied for the extraction of the conditioning factors and predictors.

Good luck in publishing your work.

Author Response

Dear Reviewer,

Thank you for your great review, feedback and kind words. Your comments and suggestions have been duly noted, and we appreciate the constructive criticism, which will help us improve our research paper. Revisions are highlighted in yellow in the attached file. Please find our response in red below:

From the title is not evident that the article is based on a literature review and no experiment is performed.

Thank you for your suggestion. We believe that a title such as " Identifying Conditioning Factors and Predictors of Conflict Likelihood for Machine Learning Models: A Literature Review" would better capture the essence of our work. We think this title communicates both the methodology of our study and its broader applicability in conflict prediction using machine learning.

Authors should provide the list of papers utlizied for the extraction of the conditioning factors and predictors.

Thank you for your insightful comments. The list of papers used for the extraction of the conditioning factors and predictors is included in Annex 1.

We hope these changes would help to address your suggestions, we appreciate your thoughtful feedback as it guides us toward improving the quality and relevance of our work. We are grateful for your invaluable feedback and look forward to further comments and suggestions.

Best regards,

Timur Obukhov and Maria Brovelli.

Author Response File: Author Response.docx

Reviewer 3 Report

The causes of armed conflict and potential conflict prediction using machine learning models are outlined in this research. This paper's topic choice is intriguing and has some practical application. The primary issues with the whole text are as follows:

1. The method is not sufficiently described in the paper. How, for instance, might conflict prediction use machine learning?

2.Armed conflict can occur for a variety of causes. Why should disputes be predicted using machine learning? What are the conventional techniques? What are the benefits and drawbacks of each? Encourage comparison and analysis.

3. Repeat the text from Line 317 to Line 325.

4. What criteria are used to determine the priorities for socioeconomic factors in Line 340-344?

5.Line 380-382:Why aren't health and infant mortality the primarily factors that lead to conflict? Does it go against the conflict's uniqueness?

Author Response

Dear Reviewer,

Thank you very much for your great review and feedback. Your comments and suggestions have been duly noted, and we appreciate the constructive criticism, which will undoubtedly help us in improving our research paper. Revisions are highlighted in yellow in the attached file. Please find our response in red below:

1. The method is not sufficiently described in the paper. How, for instance, might conflict prediction use machine learning?

Thank you for your feedback. We have taken your comment into account and expanded on our methodology section. We included a more detailed description of how machine learning can be applied to predict conflicts. We have outlined the various stages of our approach, from data preprocessing to the validation of machine learning models. In the methodology section, we have discussed how we handled unstructured data, split our data set for training and validation purposes, utilized various machine learning algorithms such as Random Forests, Neural Networks, and Support Vector Machines, and validated these models to ensure their accuracy and reliability. We believe that this expanded explanation provides a clearer understanding of how machine learning can be applied in the context of conflict prediction. Please refer to lines 119 to 148 in the revised article.

2.Armed conflict can occur for a variety of causes. Why should disputes be predicted using machine learning? What are the conventional techniques? What are the benefits and drawbacks of each? Encourage comparison and analysis.

Traditional methods of conflict analysis rely on qualitative analysis and expert judgment. These conventional methods, while valuable, can be limited by human bias, the human capacity to process vast amounts of information, and the difficulty in capturing and assessing complex patterns between multiple factors. In contrast, machine learning techniques have displayed potential for predicting conflicts due to their ability to handle large volumes of data and recognize intricate patterns across multiple variables. However, machine learning has its own limitations. These include the risk of overfitting, quality and availability of the data used, significant efforts for data pre-processing and data related to local specificities in the area of conflict. Although in our paper we are not focusing on the traditional ways and methods to analyze and predict conflicts, we strongly encourage researchers to apply a hybrid approach, i.e., to conduct research on the specificities of the geographical area, narratives related to the situation or a conflict and based on these insights identify a list of condition factors to apply in the process of developing a machine learning models. Please refer to lines 59 to 85 and lines 275 to 315 in the revised article.

3. Repeat the text from Line 317 to Line 325.

Thank you very much for pointing at it. It has been resolved.

4. What criteria are used to determine the priorities for socioeconomic factors in Line 340-344? Starting from line 201 (new)

The priorities for socioeconomic factors are determined based on their prevalence in the literature and the number of citations using the min-max normalization method. As mentioned in lines 275-315, “it is understood that the frequency of citations and the number of times a certain conditioning factor has been used in previous research does not automatically reflect the importance of a conditioning factor. This importance is often tied to specific local conditions, with varying factors and combinations differing significantly across geographical regions.”  Please refer to lines 86-95, and lines 275-315

5.Line 380-382:Why aren't health and infant mortality the primarily factors that lead to conflict? Does it go against the conflict's uniqueness?

The factors leading to conflict can be highly situation-specific, and not every factor is relevant in all situations and geographical locations. Thus, while poor health and high infant mortality can contribute to conflict, they are rarely a reason to onset of a conflict by themselves. It is often a combination of other factors, such as poor governance or economic struggle.

We hope these changes would help to address your suggestions, we appreciate your thoughtful feedback as it guides us toward improving the quality and relevance of our work. We are grateful for your invaluable feedback and look forward to further comments and suggestions.

Best regards,

Timur Obukhov and Maria Brovelli.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Thanks for your revision, and now it looks great to be published.

Reviewer 3 Report

From the prior comments, it appears that the authors improved and corrected the paper as well as included some justifiable arguments and supplements.

Congratulations to the authors.