Decision Tree-Based Federated Learning: A Survey

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

In this paper, the authors examine recent efforts to integrate federated learning and decision tree technologies and discuss communication efficiency problems in federated decision tree models. This provides theoretical support for the engineering of federated learning with decision trees as the underlying training model. The authors have done a good job of presenting their research clearly and concisely, yet there are several weaknesses in the manuscript that need attention.

1.      This paper is well-written and makes a valuable contribution to the topic of “ The technical integration of federated learning and decision trees”, yet it lacks a detailed introductory exploration of related survey work about this topic. The authors should consider completing this introductory.

2.      The paper is well-organized and easy to follow, and the authors’ conclusions are well-supported by the data. Spite the insight, there are certain weaknesses that, if addressed, would elevate its professionalism.

(1)   As written in lines 100 to 103, “We note that previous research rarely involved the integration of federated learning and decision trees, ……Therefore, the focus of this survey is the technical integration of federated learning and decision trees. ” Research of the topic was rare doesn’t mean to do the research is meaning.

(2)   As written in lines 429 to 431, “Current research on decision trees in federated learning is more focused on VFL. We speculate that under the horizontal setting, different splitting methods are difficult to aggregate to form a global model. ” Current research is more focused on VFL, can’t say directly that HFL is difficult to aggregate to form a global model.

The paper makes a valuable contribution to the field, and there is a recommendation to either integrate contributions 2 and 3, considering that communication efficiency is typically encompassed within overall performance, or alternatively, reorganize these contributions for a more coherent presentation

Comments on the Quality of English Language

In this paper, the authors examine recent efforts to integrate federated learning and decision tree technologies and discuss communication efficiency problems in federated decision tree models. This provides theoretical support for the engineering of federated learning with decision trees as the underlying training model. The authors have done a good job of presenting their research clearly and concisely, yet there are several weaknesses in the manuscript that need attention.

1.      This paper is well-written and makes a valuable contribution to the topic of “ The technical integration of federated learning and decision trees”, yet it lacks a detailed introductory exploration of related survey work about this topic. The authors should consider completing this introductory.

2.      The paper is well-organized and easy to follow, and the authors’ conclusions are well-supported by the data. Spite the insight, there are certain weaknesses that, if addressed, would elevate its professionalism.

(1)   As written in lines 100 to 103, “We note that previous research rarely involved the integration of federated learning and decision trees, ……Therefore, the focus of this survey is the technical integration of federated learning and decision trees. ” Research of the topic was rare doesn’t mean to do the research is meaning.

(2)   As written in lines 429 to 431, “Current research on decision trees in federated learning is more focused on VFL. We speculate that under the horizontal setting, different splitting methods are difficult to aggregate to form a global model. ” Current research is more focused on VFL, can’t say directly that HFL is difficult to aggregate to form a global model.

3. The paper makes a valuable contribution to the field, and there is a recommendation to either integrate contributions 2 and 3, considering that communication efficiency is typically encompassed within overall performance, or alternatively, reorganize these contributions for a more coherent presentation

Author Response

To Editor:

We feel great thanks for your professional review work on our article Decision Tree-based Federated Learning: A Survey. As you are concerned, there are several problems that need to be addressed. According to your nice suggestions, we have made extensive corrections to our previous draft, the detailed corrections are listed below.

We first made modifications to the language expression issues in the article. For example, the reviewer pointed out that the current research is more focused on VFL, and it cannot be directly stated that HFL is challenging to aggregate into a global model. Through our investigation, in a horizontal setup, due to different participants using different features to split intermediate nodes, it is difficult to aggregate directly through model parameters, as in neural networks. In contrast, in VFL, it is easier to synchronize layer by layer among parties. We have also made modifications to other issues raised by the reviewer.

We have restructured the article and added some content to make it more consistent and coherent overall. We did not find content that closely resembles our work regarding the reviewer's suggestion to add relevant research. We have already mentioned research related to federated learning and decision tree models in the introduction section.

If there are any other modifications we could make, we would like very much to modify them and we really appreciate your help.

Reviewer 2 Report

Comments and Suggestions for Authors

The article is a survey about Decision Tree-based Federated Learning methods. They focused on Federated Learning, Machine Learning, Decision Trees, Privacy Protection, and Communication Efficiency aspects. It needs a major revision. This is an interesting work and adds some new information to the existing research. I appreciate their efforts, but the paper still needs more quality. My comments for the further improvement are:

1- The abstract and conclusion should be more informative.

2- More Privacy Protection and Communication Efficiency analysis is required.

3- Lack of comments and in-depth discussions.

4- The future directions should be solid problems that can be addressed by the researchers. These future directions are based on literature and overall challenges.

5- Considering the current polls on the same subject, readers can see how the work substantially contributes. The authors overlook it. Add a subsection named "Comparison with Surveys on Decision Tree-based Federated Learning" to section one. As I searched, there is not a survey article about "Decision Tree-based Federated Learning". You can cite some surveys that cover a part of your subject. For instance, a decision tree classifier is a detailed survey on federated learning. You should draw a table and highlight which aspects the former surveys did not cover. Please find more surveys by yourself.

6- The readers need to figure out what the motivation to write the survey is. What is the need? First, ask yourself what is necessary to write a survey article. Please remember that. I recommend the authors define the scope and write an in-depth survey. Look at the current survey papers that the prestigious venues have published. Look at their motivation. 

7- Please add a section named discussion before the conclusion section. In the survey articles, it is a need. Please talk about some aspects that your research can be used in them. For instance, first, as the title of the journal is blockchain, you can discuss the subject of your survey for blockchain. Please cite some works about it. Second, as federated learning and machine learning are useful in SDN-based IoT-Fog networks, you can talk about the subject of your survey for them. Cite articles about them. For instance, cite surveys about "SDN perspective IoT-Fog security". You can discuss how we can implement privacy for federated learning methods for Fog, SDN or Blockchain. I suggest some articles that you can cite. For the third aspect, I leave the floor to the authors.

7-1 "A review on federated learning and machine learning approaches: Categorization, application areas, and blockchain technology." MDPI, 2022

7-2 "Achieving Verifiable Decision Tree Prediction on Hybrid Blockchains." MDPI, 2023

7-3 "Machine-learning techniques for predicting phishing attacks in blockchain networks: A comparative study." MDPI, 2023

7-4 "Blockchain-Empowered Federated Learning Through Model and Feature Calibration." IEEE Transactions, 2023

7-5 "Federated Learning Inspired Low-Complexity Intrusion Detection and Classification Technique for SDN-Based Industrial CPS." IEEE Transactions, 2023

7-6 "An SDN perspective IoT-Fog security: A survey." Elsevier, 2023

7-7 "Low rate DDoS detection using weighted federated learning in SDN control plane in IoT network ", MDPI, 2023

7-8 "Towards a machine learning-based framework for DDOS attack detection in software-defined IoT (SD-IoT) networks." Elsevier, 2023

Fingers crossed.

Author Response

To Editor:

Thank you for your letter and the constructive comments on this article in your busy schedule. We have carefully read the comments that you have given us, and have discussed and revised each of these issues. In addition.we have resubmitted a new manuscript in the revised state, if there are any incorrect answers or questions in the manuscript, please do not hesitate to let us know.

First, We have modified the content of the abstract and conclusion to better reflect the overall content of the article. Next, addressing the reviewer's feedback regarding the lack of comments and in-depth discussion, we have revised and enhanced the summary after each chapter. Additionally, we have introduced a Discussion section before the Conclusion, where we discuss the application of federated decision trees in scenarios such as blockchain and the Internet of Things. We also provide insights into future research directions.

The motivation behind our work is that while there are numerous research proposals combining federated learning with decision trees, there is scarce systematic integration of these works in the existing literature. Most research tends to focus on federated learning techniques using neural networks as the underlying model. We aim for readers to gain a quick understanding of federated decision tree technology through this article. In order to align with the theme of the journal, we have added content related to blockchain. Additionally, based on the feedback, we have incorporated relevant references.

If there are any other modifications we could make, we would like very much to modify them and we really appreciate your help.

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The paper is ok.