Improving Real-Time Detection of Abnormal Traffic Using MobileNetV3 in a Cloud Environment
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThis paper proposes an improved lightweight real-time detection model, named as IM-MobileNetV3, which has higher precision compared with the existing MobileNetV3 and MobileNet-Small model. The authors propose four processes for converting abnormal traffic data features into images, and introduce the ECA attention module, which achieves nice performance improvement. The overall idea of this paper is certain novel, and the description has a good logical structure. However, there are still some formatting issues as follow.
- In the keyword section, some keywords should not be bolded.
- In the introduction section, no explanation of this paper's structure arrangement is provided at the end.
- In the proposed work section, when converting the original abnormal traffic data into image features, it will be beneficial to provide some examples of the original abnormal traffic data.
- Algorithm 1 and Algorithm 2 have poor visual clarity and do not have proper indentation for the statements.
Good.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThis paper conducts research on the real-time detection technology of abnormal traffic in the cloud environment, and the research has strong practical value. However, the following problems still exist:
There are many problems in the format of the thesis. The writing formats of all author names are not uniform. For example, yue Zhao. The letters in the formula summary are not written uniformly with the corresponding letters in the main text. Some letters are in italics, while others are not. Furthermore, the font sizes of the letters in the formulas before and after the full text are also inconsistent.
2. To verify the excellent performance of the proposed model, the paper presents relevant comparative experiments. However, the presented comparison model is not the latest or mainstream one. How does the author consider this issue? Please provide an explanation. Supplement the experiment if possible.
3. The conclusion part of the paper lacks analysis and summary. How do the lightweight and high precision mentioned in the text correspond to the conclusion?
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThe proposed paper proposed the im-mobilienet v3 model to detect abnormal traffic occurring in a cloud environment in real time. It is a study that converts traffic data into rgb images and secures accuracy and real-time performance at the same time through eca attention mechanism and structural optimization.
However, there are some problems as follows.
1. Although it demonstrates model performance, it does not match the paper emphasizing real-world applicability due to lack of deployment cases or evaluation in terms of latency overhead and bandwidth throughput on actual cloud platforms (aws, azure,..).
2. Since the experiment is limited to one public dataset of cic-ids 2018, there is a lack of discussion on the application results or generalization performance to other latest abnormal traffic datasets (such as data after nsl-kdd). Since the traffic environment varies by region, industry, and time, verification using multiple datasets is necessary.
Accordingly, the following improvements are needed.
1. It is necessary to describe the key factors that contributed to the performance improvement through ablation experiments for each component of the model (eca application, 5x5 conv introduction, gap+gmp combination, etc.).
2. cic-ids2018, it is necessary to conduct extended experiments that demonstrate the versatility and robustness of the proposed model using datasets such as recent bot-iot, ton_iot, and unsw-nb15.
Comments on the Quality of English Language.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
