Core Classifier Algorithm: A Hybrid Classification Algorithm Based on Class Core and Clustering

Round 1

Reviewer 1 Report

This paper combines KNN algorithm with an unsupervised learning partitioning algorithm (K-means) to avoid the unrepresentative Cores of the clusters. The manuscript is not well-prepared. For instance, (1) all figures are not clear,(2) many equations are unreadable, (3) the format of tables are confused. So, I suggest the paper to re-write and re-organize.

Author Response

Response for reviewer1

Thank you for your valuable comments and suggestions. Your comments and suggestions are a valuable opportunity to further improve the quality of this work in terms of technical content, novelty and the quality of literary presentation. In the following section, the changes that have been introduced are presented as a result of careful inspection of the points in the review.

 

 

This paper combines KNN algorithm with an unsupervised learning partitioning algorithm (K-means) to avoid the unrepresentative Cores of the clusters. The manuscript is not well-prepared. For instance, (1) all figures are not clear?

I would like to thank you for your meaningful and helpful notes, I have edited the figure and contained a description part for each figure and I have used better resolution for the flowchart as well. you may look at lines ( 205,226,314,322).

(2) many equations are unreadable?

I have rewritten the equations. you may look at lines ( 249,256,262)

 (3) the format of tables is confusing.

I have done this part by adding descriptions for the tables. 

(4)- I suggest the paper to re-write and re-organize.

I have done proofreading for the whole paper and it's been changed into your request.

 

Author Response File: Author Response.docx

Reviewer 2 Report

The paper entitled “Core Classifier Algorithm: A Hybrid Classification Algorithm Based on Class Core and Clustering” proposes a hybrid classification framework that combines both k-means and K-NN to find the core group revealing characteristics and features of the corresponding class, thereby allowing for detecting outliers in and overlapping between groups. The proposed algorithm so-called Core Classifier Algorithm (CCA) is designed and tested on five datasets to compare the performances of CCA with other algorithms in terms of accuracy.

 

In general, the topic invested in this paper is interesting and meets the scope of the AS journal. The paper is well written, and the results seem to be reasonable. I like the way that authors combine unsupervised and supervised learning in the proposed framework. The authors should revise the paper to further improve its quality before I vote for an acceptance. My comments are as follows

 

- In the Introduction, briefly introduce the main motivations for the design of the proposed algorithm. Also, move the contributions described on page 9 to this section and discuss real-life applications of CCA.

- In section 3, put a table of notations used in the paper.

- Figures 1, 2, 3, 4 are not in a good shape. Authors should revise the figures to make the text clearer, copiable, and not be broken when zooming out.

- A lot of typos in section 3.1, the authors need to carefully proofread the paper to fix all typos and grammar mistakes.

- Authors need to reform the pseudo-code of CCA as an Algorithm. Authors can follow a template like this [https://sharelatex-wiki-cdn-671420.c.cdn77.org/learn-scripts/images/3/38/Algorithm2e-11.png]

- In section 4 Table 2, I suggest authors discuss several methods for estimating the optimal number of clusters such as using the Silhouette coefficient. Authors can refer to this paper [https://doi.org/10.1007/978-981-15-1209-4_1] in the discussion.

- In section 4.1, I suggest authors use another metric as F1 scores to validate the performance of different methods. Also, put the std (+-) for those scores.

- In section 5, discuss the problem of missing values and how to extend the proposed algorithm for such problems. Authors refer to these papers [https://doi.org/10.1016/j.ins.2021.04.076]  [https://doi.org/10.3390/sym14010060] and in the discussion.

 

Author Response

Response for reviewer 2

Thank you for your valuable comments and suggestions. Your comments and suggestions are a valuable opportunity to further improve the quality of this work in terms of technical content, novelty, and the quality of literary presentation. In the following section, the changes that have been introduced are presented as a result of careful inspection of the points in the review.

- In the Introduction, briefly introduce the main motivations for the design of the proposed algorithm. Also, move the contributions described on page 9 to this section and discuss real-life applications of CCA.?

I have worked on your notes and I have edited the introduction. and in line 105 you can find a good description for the algorithm.

- In section 3, put a table of notations used in the paper.?

I have used the table of notation and symbols .you may look at line (290)

- Figures 1, 2, 3, 4 are not in a good shape. Authors should revise the figures to make the text clearer, copiable, and not be broken when zooming out.

I have changed the figures. you may look at lines (205, 226, 314, 322).

- A lot of typos in section 3.1, the authors need to carefully proofread the paper to fix all typos and grammar mistakes.

yes, I have been after all this article and I have proofread it.

- Authors need to reform the pseudo-code of CCA as an Algorithm. Authors can follow a template like this [https://sharelatex-wiki-cdn-671420.c.cdn77.org/learn-scripts/images/3/38/Algorithm2e-11.png]

Exactly I have changed the pseudo-code formate as it’s appeared in the link. you may look at the line (287).

- In section 4 Table 2, I suggest authors discuss several methods for estimating the optimal number of clusters such as using the Silhouette coefficient. Authors can refer to this paper [https://doi.org/10.1007/978-981-15-1209-4_1] in the discussion.

I have given attention to your very nice response and I have mentioned it.

- In section 4.1, I suggest authors use another metric as F1 scores to validate the performance of different methods. Also, put the std (+-) for those scores.

the matrix has been implemented and it’s written. you may look at line (417).

- In section 5, discuss the problem of missing values and how to extend the proposed algorithm for such problems. Authors refer to these papers [https://doi.org/10.1016/j.ins.2021.04.076]  [https://doi.org/10.3390/sym14010060] and in the discussion.

thank you for your response notes being taken. you may look at line 464.

Author Response File: Author Response.docx

Reviewer 3 Report

The paper “Core Classifier Algorithm: A Hybrid Classification Algorithm 2 Based on Class Core and Clustering” introduces a new hybrid classification algorithm CNN. The article is divided into five sections. Section introduction mainly describes the field of machine learning and different classification algorithms. The section literature review describes some algorithms and some classification problems. Sections The proposed algorithm and Experimental analysis and the results describe a novel approach. The paper is concluded with a Conclusion and future work.
PROS: 
It addresses an important area of classification in the field of machine learning.
CONS: 
The article does not meet the minimum criteria of a professional article. It is written superficially with many typographical and design errors. Some pictures contain descriptions; others do not. The report does not contain all descriptions of the characters used.
Wrong or missing citation statements. For example, in line 132 authors described that they showed the accuracy of SVM and CNN…, which is not described in the paper.
Mathematical equations are written opaquely and deficiently. When describing the algorithm, we should first acidify the quality pseudocode of the proposed method and, if necessary, justify the individual steps or supplement with formal mathematical equations. The images are hard to read and repetitive (Figures 3 and 4). The paper is written in the description of the algorithm and not in the chapter on analysis and results.
The tables are vague and deficient.
It is impossible to assess whether the experiments were performed correctly as a reviewer. When proposing a new method in 2022, it is right to attach a link to its basic implementation. It is also right to give the implementation of the whole experiment, which allows repeatability of the results.
If I understand Table 3, the RF method is better than the suggested CCA in all experiments.

Author Response

Response for reviewers 3

Thank you for your valuable comments and suggestions. Your comments and suggestions are a valuable opportunity to further improve the quality of this work in terms of technical content, novelty, and the quality of literary presentation. In the following section, the changes that have been introduced are presented as a result of careful inspection of the points in the review.

Wrong or missing citation statements. For example, in line 132 authors described that they showed the accuracy of SVM and CNN…, which is not described in the paper.

I have taken your response and I have gotten the paper which is under citation 7 which has accuracy and the information you asked for.

Mathematical equations are written opaquely and deficiently.

yes, I have changed the equation format where it was sent mistakenly. you may look at the lines (249, 256, 262).

 When describing the algorithm, we should first acidify the quality pseudocode of the proposed method and, if necessary, justify the individual steps or supplement with formal mathematical equations

the pseudocode has been written in another format. you may find the information on lines (287)

. The images are hard to read and repetitive (Figures 3 and 4).

I am very thankful for your nice and meaningful comments and suggestions and figures have changed. you may look at lines (314, 322).

 The paper is written in the description of the algorithm and not in the chapter on analysis and results.
The tables are vague and deficient.

I have taken your notes and I have changed the description of the table and the table has been changed.

It is impossible to assess whether the experiments were performed correctly as a reviewer. When proposing a new method in 2022, it is right to attach a link to its basic implementation. It is also right to give the implementation of the whole experiment, which allows repeatability of the results.

I do appreciate your note,  the experiments have been attached and uploaded in the githuble and you can find the resource in the hyperlink below

https://github.com/AbdalraoufAlarbi79/CCA_algorithm.git

 
If I understand Table 3, the RF method is better than the suggested CCA in all experiments.

Yes, you understood the table and it contains 3 other algorithms and the performance varies. CCA algorithm has shown a good result as the others but not the best.

Author Response File: Author Response.docx

Reviewer 4 Report

1. Literature review needs to be reviewed and expanded and show the limitation of other work clearly. Also, the manuscript could be substantially improved by relying and citing more on recent literature.
2. The potential utilization and limitations of the results should be discussed in detail.
3. All equations required reviewed and enhanced and rewritten in a clear manner.
4. The description of evaluation metrics is not presented in the article
5. What is the computational complexity of the proposed solution? Although it is shown that the solution converges within a limited number of iterations, if the computational complexity of each iteration is high, this renders the solution to be computationally infeasible. Hence, the authors are encouraged to discuss the computational complexity of their proposed solution.
6. The authors are not making a real effort in producing the figures. All figures are not represented and bad resolution especially figure 4.
7. Having a table that summarizes the variables, sets, and notations will facilitate the reading of the paper. Please add a table with all the variables and sets used in the system model.
8. In the conclusion section, the limitations of this study suggested improvements of this work and future directions should be highlighted.

Author Response

Response for reviewer 4

Thank you for your valuable comments and suggestions. Your comments and suggestions are a valuable opportunity to further improve the quality of this work in terms of technical content, novelty, and the quality of literary presentation. In the following section, the changes that have been introduced are presented as a result of careful inspection of the points in the review.

1. Literature review needs to be reviewed and expanded and show the limitation of other work clearly. Also, the manuscript could be substantially improved by relying and citing more on recent literature.

 

thank you for your nice and beneficial response, I have reviewed the Literature and it has been edited.

 

1. The potential utilization and limitations of the results should be discussed in detail.

 

 

thank you for the information you have given the limitation has been taken. you can look at the lines(467).

 

 

1. All equations required reviewed and enhanced and rewritten in a clear manner.

 

it has been reviewed and edited. you may look at lines ( 249, 256, 262).

 

1. The description of evaluation metrics is not presented in the article

 

I have used the matrix thank you for your useful notes. you may look at line(417).

 

1. What is the computational complexity of the proposed solution? Although it is shown that the solution converges within a limited number of iterations, if the computational complexity of each iteration is high, this renders the solution to be computationally infeasible. Hence, the authors are encouraged to discuss the computational complexity of their proposed solution.

 

thank you for your kind and meaningful comments, your response has been taken and in section 3.3 line 297 and section 4.2 line 439 you may find the answers.

 

1. The authors are not making a real effort in producing the figures. All figures are not represented and bad resolution especially figure 4.

 

I have changed the resolution and the figures have been edited.

 

1. Having a table that summarizes the variables, sets, and notations will facilitate the reading of the paper. Please add a table with all the variables and sets used in the system model.

 

I have added the table for notations, you may look at line 290

 

1. In the conclusion section, the limitations of this study suggested improvements of this work, and future directions should be highlighted.

 

ok, I have reviewed the conclusion and added the limitation part. you may find the information on line 467.

 

Author Response File: Author Response.docx

Reviewer 5 Report

1. In the abstract, the spelling of Machine is not correct.
2. 100% accuracy means overfitting. The author should check the model during training and testing.
3. Reference numbering is not correct. References are not formatted properly.
4. Abbreviations are not defined. e.g., BPNN, RBFN, RF, SVM, NN, etc. Define all the abbreviations.
5. Figure 1 and Figure 2 are blurred. Improve the figure quality.
6. In algorithm: Line number and end line are missing.
7. From figure 4 (flow chart) it seems that the algorithm is not complete. The algorithm just classifies the data. Whereas in figure 4 data set was split into training and testing. In figure 4 after classification accuracy is stored. These all things are missing in the algorithm.
8. There is no need of figure 3. It is part of figure 4.
9. Why contributions are written before section 4. Move contributions at end of section 1, before the organization of the paper.
10. In figure 4, “select the”; the text is not complete
11. The paper should be read carefully. At many places, the full stop is missing. Some sentences' starting letter is small. It should be capital. In some sentences, words first letter is capital, which should be small.
12. There are many writing issues in this paper. The presentation of this work is very poor. The authors are suggested to read some high-impact factor journal papers to improve the writing quality. 

Author Response

Response for reviewer5 (Round 1)

We thank the reviewer for their comments because they are very valuable for improving the quality of our work in terms of technical contents, novelty and overall text quality. In the following section, we have answered all the points one by one and have also introduced the necessary changes in the original article.

1. in the abstract, the spelling of Machine is not correct.

 

Thank you for your response, it has been corrected.

 

1. 100% accuracy means overfitting. The author should check the model during training and testing.

 

The reason for 100% accuracy is due to the nature of the dataset. 100% accuracy was obtained only with synthetic datasets which represents an ideal scenario normally not found in the real world.

 

1. Reference numbering is not correct. References are not formatted properly.

 

I have gone through the references and I have made all the necessary changes.

 

1. Abbreviations are not defined. e.g., BPNN, RBFN, RF, SVM, NN, etc. Define all the abbreviations.

 

Thank you for noticing this deficiency. To address this problem, I have created a whole table of abbreviations on line 291.

 

1. Figure 1 and Figure 2 are blurred. Improve the figure quality.

 

The quality of the figures has been improved significantly.

 

1. In algorithm: Line number and end line are missing.

 

The necessary changes have been incorporated in the text.

 

1. From figure 4 (flow chart) it seems that the algorithm is not complete. The algorithm just classifies the data. Whereas in figure 4 data set was split into training and testing. In figure 4 after classification accuracy is stored. These all things are missing in the algorithm.

 

I have made the necessary changes

 

1. There is no need of figure 3. It is part of figure 4.

 

Yes, it is a part of figure 4 and I have used the figure to provide extra information about the dataset.

 

1. Why contributions are written before section 4. Move contributions at end of section 1, before the organization of the paper.

 

This was done in accordance to the template of the journal

 

1. In figure 4, “select the”; the text is not complete

 

The text of figure 4 has been corrected.

 

1. The paper should be read carefully. At many places, the full stop is missing. Some sentences' starting letter is small. It should be capital. In some sentences, words first letter is capital, which should be small.

 

The paper was proofread once again from the beginning and I have tried my best to remove grammar, spelling and punctuation mistakes to the best of my ability.

 

1. There are many writing issues in this paper. The presentation of this work is very poor. The authors are suggested to read some high-impact factor journal papers to improve the writing quality. 

 

The paper was proofread once again from the beginning and I have tried my best to remove grammar, spelling and punctuation mistakes to the best of my ability.

 

Round 2

Reviewer 1 Report

This paper combines KNN algorithm with an unsupervised learning partitioning algorithm (K-means) to avoid the unrepresentative Cores of the clusters. 

1. The format of tables is confusing.
2. The contribution need to be clarified.

3. Some learning-based works are missed, [1] Deep-IRTarget: An Automatic Target Detector in Infrared Imagery using Dual-domain Feature Extraction and Allocation, [2] Deep-learning-based burned area mapping using the synergy of Sentinel-1&2 data. [3]Graph-based few-shot learning with transformed feature propagation and optimal class allocation

4. The experiments are not sufficient.

Author Response

Response for reviewer1 ( Round 2)

We thank the reviewer for their comments because they are very valuable for improving the quality of our work in terms of technical contents, novelty and overall text quality. In the following section, we have answered all the points one by one and have also introduced the necessary changes in the original article.

1.The format of tables is confusing.

 

I have followed the paper template format and the tables have been edited accordingly.

 

2. The contribution need to be clarified.

 

The contribution has also been clarified in line 381.

 

3. 3. Some learning-based works are missed, [1] Deep-IRTarget: An Automatic Target Detector in Infrared Imagery using Dual-domain Feature Extraction and Allocation, [2] Deep-learning-based burned area mapping using the synergy of Sentinel-1&2 data. [3]Graph-based few-shot learning with transformed feature propagation and optimal class allocation.

These Works have been checked and included in the text in lines 127 have checked the meaning and you can look at lines 59 ans 136.

4. The experiments are not sufficient.

We appreciate this, since including the code and datasets is important these days. All the code, explanations, the datasets used and implementation are available on github repo acccessible through the link below.

AbdalraoufAlarbi79/CCA_algorithm: classification algorithm (github.com)

 

 

Reviewer 2 Report

I have checked this revision. The authors have improved the manuscript based on my comments and suggestions. Thus, I vote for an acceptance.

Author Response

I would like to thank you for your response and your good support which will give the right to publish our article in applied sciences.

Reviewer 3 Report

General comments:

• pseudocode is not a table, for pseudocode standard formatting in naming is needed ...
• for flowchart check the meaning of standard flowchart symbols (input/output->process step...)
• if you suggest a new algorithm, also provide source code with experiment... This is the only correct way to be able to replicate your results.  

Author Response

Response for reviewer3 ( Round 2)

 

We thank the reviewer for their comments because they are very valuable for improving the quality of our work in terms of technical contents, novelty and overall text quality. In the following section, we have answered all the points one by one and have also introduced the necessary changes in the original article.

• pseudocode is not a table, for pseudocode standard formatting in naming is needed ...

 

This point has been addressed in the article

 

• for flowchart check the meaning of standard flowchart symbols (input/output->process step...)

 

This point has been addressed in the article. Please look at line 332.

 

• if you suggest a new algorithm, also provide source code with experiment... This is the only correct way to be able to replicate your results.  

We appreciate this, since including the code and datasets is important these days. All the code, explanations, the datasets used and implementation are available on github repo acccessible through the link below.

AbdalraoufAlarbi79/CCA_algorithm: classification algorithm (github.com)

Reviewer 5 Report

The authors have addressed the comments to the best of my knowledge. The paper can be accepted. 

Author Response

I would like to thank you for your response and your support which will give the right to publish my article in this Apply Sciences.

Round 3

Reviewer 1 Report

The author addressed all of my concered.

Accept!

Author Response

I would like to thank you for your response and your support which will give the right to publish my article in this Apply Sciences.