Application of Natural Language Processing and Machine Learning Boosted with Swarm Intelligence for Spam Email Filtering

Round 1

Reviewer 1 Report

31. Please double check if the cost of data breach is correctly quoted

88. Author mentioned neurons here but XGBoost is not neural network. Please elaborate.

285. 100% mean value for accuracy, precision, recall and F1-measure sounds too good to be true. Please elaborate.

328. L2-norm loss function is also known as LSE. It's confusing to introduce L1/L2 regularization first, and then describe the L2-norm loss function. They are both called L2 in the paper but are different concepts.

 341/342, not very clear

350 what tree structure? Please elaborate

352 eq5 is not Taylor expansion, eq7 is the Taylor expansion of eq5. Please rephrase

360 Hyper-parameters for XGBoost are not introduced before. Please introduce them first.

380. Can you explain what individuals X mean here? Is it the vector of parameters?

721 no point to use 10^0

Author Response

I attach the reply to the reviewer's comments and the manuscript using track changes.

Author Response File: Author Response.pdf

Reviewer 2 Report

The authors present a machine learning approach to advance the spam detection. This is a useful application for all email users. The authors claim that their approach much advance than the existing approaches. The authors have presented a detailed information on the various machine learning methods applied for spam detection and provided the limitations of these existing approaches. This gives an idea to the user why an advanced method for spam detection is necessary. The authors have presented their approach very clearly and carried out a regress evaluation.

I recommend accepting the paper. I have no comments.   

Author Response

We thank the anonymous Reviewer for having carefully examined the manuscript and for the appreciation.

Reviewer 3 Report

I reviewed the paper entitled “Application of Natural Language Processing and Machine Learning Boosted with Swarm Intelligence for Spam Email Filtering,” which introduces an improved version of the sine-cosine algorithm for addressing parameter tuning of machine learning algorithms for spam email detection. In my overall appreciation, the paper has merit because: (a) it is well written and directed to the target audience of Mathematics; (b) the topic is interesting and challenging; (c) the results are well supported; and (d) the references are up-to-date and adequate.

I have the following remarks for the authors:

(a) In Line 160, the paper reads: “it has been introduced in 2004”; however, it should read “it was introduced in 2004”.

(b) In Section 2, it is tough to keep in mind the features of all state-of-the-art studies in comparison with yours. Please, add a comparative table. Here, the novel features of your proposal should be emphasized.

(c) In Line 318, there is a typo in the formula number.

(d) In Lines 320–324, some information about the parameters is presented. However, this information is not connected with the preceding and following paragraphs. Please, revise.

(e) Please, define all variables of Eqs. (5) and (6).

(f) In Line 447, please, rewrite the first sentence (it isn’t clear).

(g) In Line 501, write “di” using math notation.

(h) In Line 506, there is a typo.

(i) In the paragraph between Eqs. (23) and (24), the notation of some math expressions is incorrect. Please, revise.

(j) In Line 533, the paper reads: “We then choose the feature set with the highest overall weight.” However, which was the criterion to define what “highest overall weight” means specifically?

(k) In Figs. 1 and 2, you should use second-level numeration to identify each subfigure.

(l) In Table 6, the results are not encouraging in the case of “English 1000”. You should also discuss these results.

Author Response

I attach the reply to the reviewer's comments and the manuscript compiled using track changes.

Author Response File: Author Response.pdf