Review History


All reviews of published articles are made public. This includes manuscript files, peer review comments, author rebuttals and revised materials. Note: This was optional for articles submitted before 13 February 2023.

Peer reviewers are encouraged (but not required) to provide their names to the authors when submitting their peer review. If they agree to provide their name, then their personal profile page will reflect a public acknowledgment that they performed a review (even if the article is rejected). If the article is accepted, then reviewers who provided their name will be associated with the article itself.

View examples of open peer review.

Summary

  • The initial submission of this article was received on June 2nd, 2025 and was peer-reviewed by 3 reviewers and the Academic Editor.
  • The Academic Editor made their initial decision on August 4th, 2025.
  • The first revision was submitted on October 2nd, 2025 and was reviewed by 1 reviewer and the Academic Editor.
  • The article was Accepted by the Academic Editor on October 15th, 2025.

Version 0.2 (accepted)

· · Academic Editor

Accept

Thanks to the aurthors for their efforts to improve the work. I believe this version addressed the concerns of the reviewers successfully. It can be accepted. Congrats!

[# PeerJ Staff Note - this decision was reviewed and approved by Mehmet Cunkas, a PeerJ Section Editor covering this Section #]

Reviewer 3 ·

Basic reporting

No Comment.

Experimental design

No Comment.

Validity of the findings

No Comment.

Additional comments

The authors have fully addressed all the revisions requested in my review. I confirm that the manuscript now meets the required standards for publication.

Version 0.1 (original submission)

· · Academic Editor

Major Revisions

This article has value, but also has some issues. Please read the comments carefully and revise it accordingly.

**PeerJ Staff Note:** Please ensure that all review, editorial, and staff comments are addressed in a response letter and that any edits or clarifications mentioned in the letter are also inserted into the revised manuscript where appropriate.

Reviewer 1 ·

Basic reporting

The English language used throughout the paper can be improved. There are some minor errors like Predicted is spelled as Predi, and a better academic tone could be used throughout.
Literature can be improved. The literature review is limited, focusing on a narrow set of studies without discussing recent advancements in uncertainty quantification.
The manuscript adheres to the standards of PeerJ standards.
Yes, Intro aaddquately introduce the subject and make it clear. Just need some proofreading and language enhancement.

Experimental design

Yes, Article content within Aims and Scope of the journal
Yes, Rigorous investigation performed to a high technical & ethical standard
Must be improved, Methods described with sufficient detail & info to replicate
Must be improved, Discussion on data preprocessing sufficient/required
Yes, adequately described evaluation methods, assessment metrics, model selection methods
Yes, sources are adequately cited

Validity of the findings

Yes, Experiments and evaluations performed satisfactorily and Conclusion identifies unresolved questions/limitations/future directions

·

Basic reporting

The paper is well-written and clearly structured, providing a logical flow from the problem statement to the proposed methodology and experimental results. The authors clearly state the motivation behind the work, and the research questions are relevant and timely, especially given the increasing reliance on AI in high-stakes decision-making like fraud detection. Figures and tables are generally well-labeled and informative. The language is clear and grammatically correct

Experimental design

The paper compares multiple deep learning architectures, with and without UQ mechanisms, including Monte Carlo Dropout, Ensemble DNN, and Ensemble Monte Carlo Dropout (EMCD). This comparative setup is commendable as it offers a balanced view of how different UQ techniques impact predictive performance and reliability. However, the rationale behind the choice of specific hyperparameters, model architectures, and ensemble sizes is not always sufficiently justified. A more detailed ablation study could strengthen the conclusions by isolating the contribution of each component

Validity of the findings

The paper addresses a crucial challenge in the domain of insurance fraud detection by exploring how uncertainty quantification (UQ) can enhance the trustworthiness of deep learning models. In assessing the validity of the findings, the study appears to adopt a rigorous methodological approach, integrating uncertainty estimation techniques such as Monte Carlo Dropout and Bayesian Neural Networks into conventional fraud detection architectures. The findings presented are largely valid within the controlled experimental setup

Additional comments

This paper is a promising contribution with significant potential impact. With further refinements and expansions in the above areas, it can serve as a strong reference for both academic and applied communities in AI-driven fraud detection.

Reviewer 3 ·

Basic reporting

The manuscript is well-written and demonstrates a high standard of academic English throughout. The abstract provides a concise and effective overview of the study’s motivation, methodology, and principal findings. The introduction establishes a robust contextual foundation, clearly articulating both the practical importance of insurance fraud detection and the growing necessity of integrating uncertainty quantification (UQ) into deep learning frameworks. The reference list is comprehensive and up to date, encompassing both foundational and recent contributions to the field, which reflects a thorough engagement with the relevant literature.

That said, the literature review would benefit from a more synthesized and critically reflective presentation. Currently, it tends to adopt a descriptive, enumerative structure that outlines prior studies without sufficiently elaborating on their methodological distinctions, comparative strengths, or inherent limitations. A more analytical approach would strengthen the scholarly rigor and better position the current work within the existing body of knowledge.

Figures and tables are clearly designed, appropriately labeled, and effectively support the narrative. The overall structure adheres to the formatting and organizational standards of *PeerJ*, and the reporting quality meets the journal’s requirements. With minor revisions to improve narrative coherence and deepen the comparative discussion—particularly in the evaluation of related work—the manuscript would gain greater clarity, logical flow, and academic impact.

Experimental design

Although the manuscript is generally well-structured and clearly articulated, certain aspects require refinement to fully align with the scholarly standards expected by the journal.

First, while the literature review is comprehensive in scope, it lacks sufficient critical synthesis and comparative analysis. The current presentation predominantly adopts a descriptive, chronological listing of prior studies, with limited discussion of their methodological shortcomings, contextual applicability, or interrelationships. To enhance scholarly depth, the authors are encouraged to reframe this section with a more analytical perspective—one that explicitly identifies key research gaps, evaluates the evolution of methodological approaches in the field, and clearly positions the present work within a coherent intellectual trajectory.

Second, the introduction effectively underscores the importance of uncertainty quantification (UQ) in deep learning models for fraud detection; however, the novelty and distinctiveness of the proposed methodology are not sufficiently articulated. The authors should more explicitly delineate how their approach advances beyond existing UQ applications in related domains. In particular, a clearer justification of the integration of SHAP (SHapley Additive exPlanations), EMCD (Expected Model Confidence Distribution), and UQ metrics—and how their synergistic use contributes to improved interpretability and reliability in fraud detection—would strengthen the perceived originality and significance of the work.

Finally, certain technical descriptions, especially those related to the uncertainty confusion matrix, are conceptually dense and may challenge reader comprehension. These sections could benefit from more concise exposition and the inclusion of visual aids or schematic illustrations to clarify complex ideas. Such enhancements would improve accessibility for a broader audience while maintaining technical rigor.

Addressing these points would substantially enhance the manuscript’s clarity, scholarly contribution, and overall impact.

Validity of the findings

No comment

Additional comments

No comment

All text and materials provided via this peer-review history page are made available under a Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.