A Novel PCA-Firefly Based XGBoost Classification Model for Intrusion Detection in Networks Using GPU

Round 1

Reviewer 1 Report

Please explain the limits of existing IDS systems and how this paper addresses the limits.

Please highlight the research value and your contribution of this work.

Meanwhile, please make it clear if this paper designs those algorithms or just uses them. In Section 2, do not list equations if they were not proposed/created by this paper.

From the evaluation results, I didn't see any significant improvement on the detection accuracy. And it is unknown if PCA+Firefly can reduce the training time. Therefore, the evaluation results do not support the authors' goal in this paper. 

 

 

Author Response

At the outset, we would like to extend our sincere thanks to Reviewer 1 for the valuable comments. Herewith the rebuttals related to the queries are addressed. 

Comment 1: Please explain the limits of existing IDS systems and how this paper addresses the limits.

Response: This is addressed in the last paragraph starting from Line 152 in the "Related work" section.

Comment 2: Please highlight the research value and your contribution of this work.

Response: This is added in the paragraph starting at Line 69 of "Introduction". 

Comment 3: Meanwhile, please make it clear if this paper designs those algorithms or just uses them. In Section 2, do not list equations if they were not proposed/created by this paper.

Response: The paper uses the algorithms and hence as per the suggestions, the equations have been removed. 

Comment 4: From the evaluation results, I didn't see any significant improvement on the detection accuracy. And it is unknown if PCA+Firefly can reduce the training time. Therefore, the evaluation results do not support the authors' goal in this paper. 

Response: Although the training time is slightly higher with respect to some of the popular algorithms as seen in Table 1 in the paper,the proposed model yields better performance and outperforms the other model in terms of Specificity and Sensitivity. Also it is observed in Table 1, that PCA+Firefly reduces training time of all the ML algorithms considered in this work. 

Reviewer 2 Report

This paper proposed a classification method for intrusion detection. The authors studied on XGBoost model based on PCA-firefly for their proposed method.

This paper just analyzed a data set related to intrusion detection from Kaggle repository. So, in section 3, the proposed method is very simple and a description of simple data analysis using convenient PCA and popular firefly algorithm. The authors must add more detail explanation of their proposed method.

I think this paper seems to have finished very urgently. So, in a large part of this paper, I found misspellings and grammatical errors. In order to solve this problem, the authors must once again check the contents of the paper and solve them. For example;

Line 154 in page 5: The authors should check as follow;

IN this work ~ -> In this work ~

Line 163 in page 5: The authors should check as follow;

~, normalize aal the ~ -> ~, normalize all the ~

In addition, the title of this paper contains the keyword “GPU”, but I did not find any explanation about it. So, the authors should explain the reason why they use it in the title.

Author Response

The authors sincerely thank the valuable comments of Reviewer 2 which are addressed herewith. Comment 1: This paper just analyzed a data set related to intrusion detection from Kaggle repository. So, in section 3, the proposed method is very simple and a description of simple data analysis using convenient PCA and popular firefly algorithm. The authors must add more detail explanation of their proposed method. Response: The detailed explanation of the proposed method has been provided in the "Proposed Methodology" section. Comment 2: I think this paper seems to have finished very urgently. So, in a large part of this paper, I found misspellings and grammatical errors. In order to solve this problem, the authors must once again check the contents of the paper and solve them. For example; Line 154 in page 5: The authors should check as follow; IN this work ~ -> In this work ~ Line 163 in page 5: The authors should check as follow; ~, normalize aal the ~ -> ~, normalize all the ~ In addition, the title of this paper contains the keyword “GPU”, but I did not find any explanation about it. So, the authors should explain the reason why they use it in the title. Response: All the typographical and grammatical errors have been rectified. The concept and explanation of GPU has been provided in Line 275 and Line 276 in the "Results and Discussion section".

Reviewer 3 Report

General:
* related work is part of the introduction.
* other approaches have to be compared to the proposed approach
* XGBoost, PCA, and Firefly should be explained in more detail and the others should be removed, since they are not part of the approach
* Please explain all variables of the equations (1), (8), (9)
* Reference for the IDS Kaggle data set is missing
* Line 148: Reference for One-Hot Encoding is missing
* Line 157: Sentence: As XGBoost has inbuilt ... -> make no sence is not an advantage
* Equation (11) should be explained in more detail. This is the crusial part of the paper
* Result figures (1-7) have to be discussed
* The result is promissing, but cost for processing is missing. I guess there the approach is a lot worse then pure ML algorithms

 

Spelling and minor corrections:
* Line 76: there is a comma and dot -> only dot
* Line 154: IN -> In
* L 163: aal -> all
* L 198: AFter -> After

Author Response

The authors sincerely thanks the Reviewer 3 for the valuable comments and the responses are addressed herewith. 

Comment 1: related work is part of the introduction.

Response: The Introduction and Related Work have been segregated in the updated manuscript. 

Comment 2: other approaches have to be compared to the proposed approach

Response: Figure 9 presents the comparison of the proposed method with the other approaches. 

Comment 3: XGBoost, PCA, and Firefly should be explained in more detail and the others should be removed, since they are not part of the approach

Response: The Background section has been modified with detailed explanation of the relevant algorithms. 

Comment 4: Please explain all variables of the equations (1), (8), (9)

Response: As per the comments received from other reviewers, some of the equations have been removed. However, the equations 8 and 9 have been updated as Equation 5 and 6 with variables clearly explained. 

Comment 5:Reference for the IDS Kaggle data set is missing

Response: The reference for the  IDS Kaggle data set has been added as Reference no. 30 in the Results and Discussion section.

Comment 6: Line 148: Reference for One-Hot Encoding is missing

Response: The reference for One-Hot Encoding has been added as Reference no. 29 in the Proposed Methodology section.

Comment 7: Line 157: Sentence: As XGBoost has inbuilt ... -> make no sense is not an advantage

Response: The sentence has been modified and updated in Line 242. 

Comment 8: Equation (11) should be explained in more detail. This is the crusial part of the paper

Response: The Equation 11 has been updated to Equation 9 in the modified version of the manuscript.

Comment 9: Result figures (1-7) have to be discussed

Response: The updated Figure numbers (4-9) have been discussed in the Results and Discussion section. 

Comment 10: The result is promissing, but cost for processing is missing. I guess there the approach is a lot worse then pure ML algorithms

Response: Table 1 has been added that highlight the cost for processing. 

Round 2

Reviewer 1 Report

Please fix the indexing error in figures.

Author Response

The authors sincerely thank the valuable comments of Reviewer 1 which are addressed herewith. 

The following are the comments and responses pertinent to Review 2:

Comment 1: Please fix the indexing error in figures.

Response: The indexing of the figures have been updated.

Reviewer 2 Report

I could not find any response to the modifications I requested the authors in the 1st review.

In addition, the sentences related to the main contributions of this work should be moved to section 4.

Author Response

The authors sincerely thank the valuable comments of Reviewer 2 which are addressed herewith.

The following are the comments and responses pertinent to Review 1 for your kind perusal. 

Comment 1: This paper just analyzed a data set related to intrusion detection from Kaggle repository. So, in section 3, the proposed method is very simple and a description of simple data analysis using convenient PCA and popular firefly algorithm. The authors must add more detail explanation of their proposed method.

Response: The detailed explanation of the proposed method has been provided in the "Proposed Methodology" section.

Comment 2: I think this paper seems to have finished very urgently. So, in a large part of this paper, I found misspellings and grammatical errors. In order to solve this problem, the authors must once again check the contents of the paper and solve them. For example; Line 154 in page 5: The authors should check as follow; IN this work ~ -> In this work ~ Line 163 in page 5: The authors should check as follow; ~, normalize aal the ~ -> ~, normalize all the ~ In addition, the title of this paper contains the keyword “GPU”, but I did not find any explanation about it. So, the authors should explain the reason why they use it in the title.

Response: All the typographical and grammatical errors have been rectified. The concept and explanation of GPU has been provided in Line 275 and Line 276 in the "Results and Discussion section".

The following are the comments and responses pertinent to Review 2:

Comment 1: In addition, the sentences related to the main contributions of this work should be moved to section 4.

Response: The main contributions have been moved to the last paragraph in Sector 4 – Proposed Methodology starting from Line 262.

Reviewer 3 Report

The approach of a Hybrid Principal Component Analysis (PCA) – Firefly based
machine learning model to classify IDS datasets is well described. The experimental results of the Hybrid PCA-Firefly algorithm for dimensional reduction showed is effectiveness.

Author Response

We would thank Reviewer 3 for the positive feedback on the manuscript submitted.

Round 3

Reviewer 2 Report

I think the authors have solved the problems noted in previous reviews. However, the authors still need to add more explanation on their proposed method. There is a lack of explanation of the proposed method compared to the research background.

Author Response

The authors would like to thank the reviewer for the valuable comments in enhancing the quality of the manuscript. The responses pertinent to Review 3 are mentioned herewith for your kind perusal. 

Comment: I think the authors have solved the problems noted in previous reviews. However, the authors still need to add more explanation on their proposed method. There is a lack of explanation of the proposed method compared to the research background.

Response: An enhanced explanation about the advantages of the proposed method is given in the Proposed Methodology section in the updated manuscript in lines 228-242.