Intelligent Unsupervised Network Traffic Classification Method Using Adversarial Training and Deep Clustering for Secure Internet of Thingsâ€
Round 1
Reviewer 1 Report
The paper is an interesting one in a very actual domain. Also there are good references and most of them are vey actual (from the last 3...4 years) and this reflect that the paper subject is in the center of today interest of scientific community.
The proposed method is combining a set of computational procedures in order to obtain a valuable classification. Some evaluation of this computational effort (not only the accuracy) may be also of practical interest of readers.
There are not issues with the English usage, but an extensive and careful proofreading at the end is welcome.
Author Response
Many thanks for your strong support and positive suggestions. We shall carefully revise this paper according to your comments and comments from other reviewers. Hope our work will make a contribution to the actual application scenarios.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Overall, a good manuscript should consider the following comments:
1. In the abstract, please consider indicating possible limitations of this work and any future steps.
2. While the introduction provides a good overview, the authors should consider:
-Elaborating on how the adversarial auto-encoder (CAAE) differs from other auto-encoders used in the field.
-Providing more details on the optimization method DC-CAAE and how it enhances the CAAE feature extraction.
-Discussing potential limitations or challenges when using the proposed method in real-world scenarios.
A deeper dive into the challenges and limitations of using auto-encoders in this context could be beneficial.
The clustering algorithm's specifics could be elaborated upon, especially the choice of distance metric as well as its implications.
The authors might consider elaborating on the specific advantages of the DC cell over other potential solutions.
A discussion on any limitations of the proposed method could be beneficial for better reading purposes.
3. Extend the determination of the number of clusters using the elbow method and silhouette coefficient method.
4. The article could benefit from more detailed insights into the training process and its implications.
5. Tests reveal a multi-classification accuracy of 92.2%, making it effective for classifying large volumes of unlabeled data in real-world situations. How about other methods?
6. In the conclusions section, please try to include the receivers of this work. How are they going to benefit from it? How are you going to communicate your research findings with them?
Minor grammatical errors could be corrected by thoroughly proofreading the manuscript.
Author Response
We would like to thank the editor for handling this manuscript and giving us the opportunity to revise the manuscript. We would also like to express our appreciation to the reviewers for providing us with valuable comments for improving the manuscript. We have carefully revised our manuscript according to the comments, which we sincerely appreciate because they helped us to improve the quality of our manuscript.
In the following, we present a point-by-point reply to the reviewers' comments. To make it easier to read, the comments from each reviewer are in blue, and the revised portions in the new manuscript are marked in yellow. We hope you find our revisions satisfactory. If there are further comments, we are pleased to address them as well.
Author Response File: Author Response.pdf
