Hybrid GA-SVM Approach for Postoperative Life Expectancy Prediction in Lung Cancer Patients

Round 1

Reviewer 1 Report

This manuscript applied an existing method, the hybrid GA-SVM algorithm, to one single existing dataset from the UCI dataset repository. By comparing accuracy, precision, recall, and F1 score with another article using the same dataset, the authors concluded that their proposed GA-SVM performed better and was suggested for detecting post-operative life expectancy in lung cancer patients. Methodology and results were written clearly enough for the reviewer to understand. However, moderate English changes are required due to weird pronouns, tenses, undefined abbreviations, and other writing problems all over the manuscript.

Some major problems are identified below:

1. What is the key innovation of the manuscript? It seems that the authors regarded their proposed GA-SVM algorithm as an innovation. However, GA-SVM has been developed for a long time. If one checks the Google Scholar ( https://scholar.google.com/scholar?hl=en&as_sdt=0%2C25&q=GA-SVM&btnG= ), there are already a bunch of GA-SVM methodology and applications. So what is the key difference between the authors’ GA-SVM algorithm and the existing GA-SVM algorithm?

2. It’s typical in the healthcare area that people use existing algorithms to better solve a medical problem. But if it’s in this case, much more medical explanations are required.

3. Only one dataset was checked for the results. The generalizability is a big issue.

4. Only 454 observations were used in modeling. The authors stated that “large enough for analysis.” However, how did they decide?

5. The results (see Table 2) were compared to Multilayer Perceptron (MLP), J48, and Naïve Bayes methods used in another article. However, this cited article was published on arXiv only, according to my search. It was not a peer-reviewed or other kind of reliable article at all. The results (also Table 2) were also compared to the Random Forest method. However, the corresponding citation [1] was just a general WHO webpage without any results about random forests. So I could not know where the baseline results came from. That is, based on the current manuscript, all baseline results compared to were not reliable, and there were no other baselines for their results in this manuscript.

Overall, the current manuscript is better than a big course project but way not enough to be a journal article.

Author Response

                                            A summary of changes

 

 

 

Hybrid GA-SVM Approach for Post-Operative Life Expectancy in the Lung Cancer Patients

Arfan Ali Nagra ^1,*, Iqra Mubarik ¹, Muhammad Mugees Asif ¹, Mohammed A. Al Ghamdi ² and Sultan H. Almotiri ²

¹   Department of Computer Science, Garrison University, Lahore 94777, Pakistan; arfan137nagra@gmail.com (A.A.N.); iqramubarik914@gmail.com (I.M); mugeesasifm@gmail.com (M.M.A.)

²   Computer Science Department, Umm Al-Qura University, Makkah City 21961, Saudi Arabia; maeghamdi@uqu.edu.sa (M.A.G.); shmotiri@uqu.edu.sa (S.H.A.)

*              Correspondence: arfan137nagra@gmail.com (A.A.N.)       

 

 

First of all, let us express our most sincere thanks to Reviewers, who has given us many helpful comments and a chance to improve our works. Frankly, speaking, if there were no your constructive comments and suggestions, it was not possible for this paper to be improved to such a new version.

The main revisions according to the Reviewers are made as follows:

 

Reply to Reviewer 1:

• What is the key innovation of the manuscript? It seems that the authors regarded their proposed GA-SVM algorithm as an innovation. However, GA-SVM has been developed for a long time. If one checks the Google Scholar ( https://scholar.google.com/scholar?hl=en&as_sdt=0%2C25&q=GA-SVM&btnG= ), there are already a bunch of GA-SVM methodology and applications. So what is the key difference between the authors’ GA-SVM algorithm and the existing GA-SVM algorithm?

Ans. Genetic has been efficiently utilized in the feature selection problem to redact of high-dimensional dataset. One of the drawback of this technique is that it cannot build the association between the features when selecting the optimal features. The possibility of selecting a subset with redundancy has been increased. To reduce these weaknesses, the Manuscript introduces a boosted genetic algorithm has been proposed for the finest selection of a feature subset from a multidimensional. The proposed approach splits the chromosome into numerous classifications for local management.  So, mutation and crossover operators has been used on stated groups to eliminate invalid chromosomes.

• It’s typical in the healthcare area that people use existing algorithms to better solve a medical problem. But if it’s in this case, much more medical explanations are required..

Ans. In the revised manuscript, in the introduction section required information about the lungs has been added.  

•  Only one dataset was checked for the results. The generalizability is a big issue.. 

Ans In the revised manuscript, Haberman dataset and appendicitis dataset has been used to solve the generalizability issue.

 

• Only 454 observations were used in modeling. The authors stated that “large enough for analysis.” However, how did they decide?

Ans. In the revised manuscript, the data set comprises 454 observations instead of the original 470, which representing the good for analysis.  Because, most of the researcher used less than 470 observation, which is mention in the reference [19]. 

    

• The results (see Table 2) were compared to Multilayer Perceptron (MLP), J48, and Naïve Bayes methods used in another article. However, this cited article was published on arXiv only, according to my search. It was not a peer-reviewed or other kind of reliable article at all. The results (also Table 2) were also compared to the Random Forest method. However, the corresponding citation [1] was just a general WHO webpage without any results about random forests. So I could not know where the baseline results came from. That is, based on the current manuscript, all baseline results compared to were not reliable, and there were no other baselines for their results in this manuscript.
• In the revised manuscript, the preprint article has been removed and research paper named Hybrid Method to Predict Postoperative Survival of Lung Cancer Using Improved SMOTE and Adaptive SVM has been added for comprehensive comparisons. Also, three comparison tables have been added with the proposed methodology”

 

 

 

Sincerely yours,

    Arfan Ali Nagra,

 

 

 

 

 

 

Author Response File: Author Response.docx

Reviewer 2 Report

Some minor english syntactical and grammatic errors should be corrected, including removal of first person perspective writing (such as in section 4.2 and 5).

Author Response

                                            A summary of changes

 

 

 

Hybrid GA-SVM Approach for Post-Operative Life Expectancy in the Lung Cancer Patients

Arfan Ali Nagra ^1,*, Iqra Mubarik ¹, Muhammad Mugees Asif ¹, Mohammed A. Al Ghamdi ² and Sultan H. Almotiri ²

¹   Department of Computer Science, Garrison University, Lahore 94777, Pakistan; arfan137nagra@gmail.com (A.A.N.); iqramubarik914@gmail.com (I.M); mugeesasifm@gmail.com (M.M.A.)

²   Computer Science Department, Umm Al-Qura University, Makkah City 21961, Saudi Arabia; maeghamdi@uqu.edu.sa (M.A.G.); shmotiri@uqu.edu.sa (S.H.A.)

*              Correspondence: arfan137nagra@gmail.com (A.A.N.)       

 

 

First of all, let us express our most sincere thanks to Reviewers, who has given us many helpful comments and a chance to improve our works. Frankly, speaking, if there were no your constructive comments and suggestions, it was not possible for this paper to be improved to such a new version.

The main revisions according to the Reviewers are made as follows:

 

Reply to Reviewer 2:

 

• Some minor english syntactical and grammatic errors should be corrected, including removal of first person perspective writing (such as in section 4.2 and 5).

 

Ans. In the updated manuscript, english syntactical and grammatical errors  have been corrected and removed the first person perspective.

 

 

 

Sincerely yours,

    Arfan Ali Nagra,

Reviewer 3 Report

Article Number：applsci-1764538

Title: Hybrid GA-SVM Approach for Post-Operative Life Expectancy in the Lung Cancer Patients

Comments to Author:

The article proposes an algorithm that combines a genetic algorithm with a support vector machine to predict postoperative survival of lung cancer patients. This work is of interest in the field of postoperative cancer prediction. In my opinion, this work is well done and it deserves to be published. However, the manuscript itself still has significant problems and would require major revisions if published. Our detailed comments are as follows：

1.   For the quantitative analysis of the prediction models, the authors used only accuracy, precision, recall, and F1 scores. These metrics cannot give a comprehensive evaluation of the model. We would like the author(s) to add two metrics, G-mean and AUC, to increase the persuasiveness of the results.

2.   The authors used the dataset produced and published by Wroclaw Thoracic Surgery Center, which is identical to Jiang's article Hybrid Method to Predict Postoperative Survival of Lung Cancer Using Improved SMOTE and Adaptive SVM. We hope that the author(s) will add a comparison experiment to illustrate the superiority of this article over Jiang's method.

3.   The manuscript is very irregular in its use of acronyms, with several acronyms used directly without explanation. There are a number of acronyms where the explanation appears in the second or third use rather than the first use.

4.   Chapter 3.1 uses the letter 'C' directly without any explanation, which makes the reader confused.

5.   Please check why the correct death prediction curve in Figure 3 exceeds the value 1 at class1, which is not reasonable. Also, please replace Figure 2 with Figure 3, their pixel values are too low making it difficult for the reader to read.

Author Response

                                            A summary of changes

 

 

 

Hybrid GA-SVM Approach for Post-Operative Life Expectancy in the Lung Cancer Patients

Arfan Ali Nagra ^1,*, Iqra Mubarik ¹, Muhammad Mugees Asif ¹, Mohammed A. Al Ghamdi ² and Sultan H. Almotiri ²

¹   Department of Computer Science, Garrison University, Lahore 94777, Pakistan; arfan137nagra@gmail.com (A.A.N.); iqramubarik914@gmail.com (I.M); mugeesasifm@gmail.com (M.M.A.)

²   Computer Science Department, Umm Al-Qura University, Makkah City 21961, Saudi Arabia; maeghamdi@uqu.edu.sa (M.A.G.); shmotiri@uqu.edu.sa (S.H.A.)

*              Correspondence: arfan137nagra@gmail.com (A.A.N.)       

 

 

First of all, let us express our most sincere thanks to Reviewers, who has given us many helpful comments and a chance to improve our works. Frankly, speaking, if there were no your constructive comments and suggestions, it was not possible for this paper to be improved to such a new version.

The main revisions according to the Reviewers are made as follows:

 

Reply to Reviewer 3:

• For the quantitative analysis of the prediction models, the authors used only accuracy, precision, recall, and F1 scores. These metrics cannot give a comprehensive evaluation of the model. We would like the author(s) to add two metrics, G-mean and AUC, to increase the persuasiveness of the results.

Ans. In the revised manuscript (applsci-1763996), comprehensive evaluation of the model like G-mean and AUC to increase the persuasiveness of the results has been added and calculated. ..

•  The authors used the dataset produced and published by Wroclaw Thoracic Surgery Center, which is identical to Jiang's article Hybrid Method to Predict Postoperative Survival of Lung Cancer Using Improved SMOTE and Adaptive SVM. We hope that the author(s) will add a comparison experiment to illustrate the superiority of this article over Jiang's method.

Ans. The research paper named Hybrid Method to Predict Postoperative Survival of Lung Cancer Using Improved SMOTE and Adaptive SVM has been considered for comparison for the purpose of experimental analysis. The three comparison tables have been added in the manuscript.

 

    

• The manuscript is very irregular in its use of acronyms, with several acronyms used directly without explanation. There are a number of acronyms where the explanation appears in the second or third use rather than the first use.

Ans.  In the updated manuscript, all the acronyms have been explained according to your instructions.

 

• Chapter 3.1 uses the letter 'C' directly without any explanation, which makes the reader confused.

Answer:  in the revised manuscript, all the acronyms have been explained according to your instructions.

 Answer: We have replaced the figures in overall manuscript with enhanced pixels.

• Please check why the correct death prediction curve in Figure 3 exceeds the value 1 at class1, which is not reasonable. Also, please replace Figure 2 with Figure 3, their pixel values are too low making it difficult for the reader to read.”

Ans. In the revised manuscript, all the figures have been updated according to your instructions.

 

 

 

Sincerely yours,

    Arfan Ali Nagra,

 

 

 

 

 

 

Round 2

Reviewer 1 Report

This revised version is much better than the original version. Even though there're still some problems but it's worth to be published to the research community for further discussion.

Author Response

                                            A summary of changes

 

 

 

Hybrid GA-SVM Approach for Post-Operative Life Expectancy in the Lung Cancer Patients

Arfan Ali Nagra ^1,*, Iqra Mubarik ¹, Muhammad Mugees Asif ¹, Mohammed A. Al Ghamdi ² and Sultan H. Almotiri ²

¹   Department of Computer Science, Garrison University, Lahore 94777, Pakistan; arfan137nagra@gmail.com (A.A.N.); iqramubarik914@gmail.com (I.M); mugeesasifm@gmail.com (M.M.A.)

²   Computer Science Department, Umm Al-Qura University, Makkah City 21961, Saudi Arabia; maeghamdi@uqu.edu.sa (M.A.G.); shmotiri@uqu.edu.sa (S.H.A.)

*              Correspondence: arfan137nagra@gmail.com (A.A.N.)       

 

 

First of all, let us express our most sincere thanks to Reviewers, who has given us many useful comments and a chance to improve our work. Frankly speaking, it is due to your constructive comments and suggestions that the paper has been improved and transformed to such a new version.

 

The main revisions are presented as follows:

 

Reply to Reviewer 1:

• What is the key innovation of the manuscript? It seems that the authors regarded their proposed GA-SVM algorithm as an innovation. However, GA-SVM has been developed for a long time. If one checks the Google Scholar ( https://scholar.google.com/scholar?hl=en&as_sdt=0%2C25&q=GA-SVM&btnG= ), there are already a bunch of GA-SVM methodology and applications. So what is the key difference between the authors’ GA-SVM algorithm and the existing GA-SVM algorithm?

 

Response: The genetic has been efficiently utilized in the feature selection problem to redirect the high-dimensional dataset. One of the drawback of this technique is that it cannot build the association between the features when selecting the optimal features. The possibility of selecting a subset with redundancy has been increased. To reduce these weaknesses, the Manuscript introduces a boosted genetic algorithm that has been proposed for the finest selection of a feature subset from a multidimensional. The proposed approach splits the chromosome into numerous classifications for local management. So, mutation and crossover operators has been used on stated groups to eliminate invalid chromosomes.

• It’s typical in the healthcare area that people use existing algorithms to better solve a medical problem. But if it’s in this case, much more medical explanations are required.

 

Response: In the introduction section of the revised manuscript, the required medical information about the lung cancer and its diagnosis has been added.  

 

• Only one dataset was checked for the results. The generalizability is a big issue. 

 

Response: In the revised manuscript, two datasets, the Haberman dataset and Appendicitis dataset have been used to solve the generalizability issue.

 

• Only 454 observations were used in modeling. The authors stated that “large enough for analysis.” However, how did they decide?

 

Response: In the revised manuscript, the data set comprises 454 observations instead of the original 470, which is considered sufficient for the initial experimental analysis as a number of researcher used even less than 470 observations. The typical example to support our motion is the research work in the reference [19]. 

    

• The results (see Table 2) were compared to Multilayer Perceptron (MLP), J48, and Naïve Bayes methods used in another article. However, this cited article was published on arXiv only, according to my search. It was not a peer-reviewed or other kind of reliable article at all. The results (also Table 2) were also compared to the Random Forest method. However, the corresponding citation [1] was just a general WHO webpage without any results about random forests. So I could not know where the baseline results came from. That is, based on the current manuscript, all baseline results compared to were not reliable, and there were no other baselines for their results in this manuscript.

 

Ans. In the revised manuscript, the preprint article has been removed. The article named Hybrid Method to Predict Postoperative Survival of Lung Cancer Using Improved SMOTE and Adaptive SVM has been added for the comprehensive comparison. Also, four new comparison tables have been added and compared with the proposed methodology in Page numbers 7 and 8.

Table 2. Comparison of existing Approaches with respect to Accuracy and F1 score.
Model	Accuracy	AUC
FPSO-SVM + SMOTE [22]	0.8792	0.8807
PSO-SVM + SMOTE [22]	0.8713	0.7602
SVM + SMOTE [24]	0.7979	0.7966
KNN + SMOTE [23]	0.7708	0.7736
GA-SVM+ SMOTE Proposed method	0.90	0.78

 

 

 

 

 

 

Table 3. Comparison of G-mean and AUC with different models on the Haberman dataset
Model	G-mean	AUC
FPSO-SVM + SMOTE [22]	0.6942	0.6813
PSO-SVM + SMOTE [22]	0.5832	0.6131
SVM+ SMOTE [24]	0	0.6096
KNN+ SMOTE [23]	0.6572	0.6649
Proposed Method	0.7897	0.6989

 

 

 

 

 

 

 

 

 

 

 

 

Table 4. Accuracy and AUC comparison for different algorithms on the Haberman dataset.
Model	Accuracy	F1 score
RANDOM FOREST [1]	0.83	0.91
FPSO-SVM + SMOTE [22]	0.6890	0.6612
PSO-SVM +SMOTE [22]	0.6435	0.5089
SVM + SMOTE [24]	0.6291	0
KNN + SMOTE [23]	0.6630	0.6545
GA-SVM+ SMOTE Proposed method	0.85	0.91

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

As shown in Table 2, the Random Forest technique has performed relatively better than other four algorithms, namely, FPSO-SVM + SMOTE, PSO-SVM +SMOTE, SVM + SMOTE and KNN + SMOTE. It has achieved 83% classification accuracy, which is better than all other classifiers. In Table 4, GA-SVM + SMOTE outperformed PSO + SMOTE, SVM + SMOTE, and KNN + SMOTE. Its Accuracy and AUC results, which are both 0.85 and 0.69, are better than those of any other classifiers. Table 5 shows that proposed method performed better than the PSO + SMOTE, SVM + SMOTE, and KNN + SMOTE.      

 

 

Table 5. Accuracy and AUC comparison for different algorithms on the appendicitis dataset.
Model	Accuracy	AUC
FPSO-SVM + SMOTE [22]	0.6890	0.6813
PSO-SVM + SMOTE [22]	0.6435	0.6131
SVM + SMOTE [24]	0.6291	0.6096
KNN + SMOTE [23]	0.6630	0.6649
GA-SVM+ SMOTE Proposed method	0.85	0.69

 

 

 

 

 

 

 

Sincerely yours,

                        Arfan Ali Nagra,

 

 

 

 

 

 

Author Response File: Author Response.docx

Reviewer 3 Report

Thanks to the authors for answering my concerns. I had read the revised manuscript. Compared with the previous version, the current had a significant improvement.

The author's reply is also very confusing and not easy to understand.

There are still many irregularities in this revised manuscript, which does not meet the publication standards of the  journal.

Author Response

                                            A summary of changes

 

 

 

Hybrid GA-SVM Approach for Post-Operative Life Expectancy in the Lung Cancer Patients

Arfan Ali Nagra ^1,*, Iqra Mubarik ¹, Muhammad Mugees Asif ¹, Mohammed A. Al Ghamdi ² and Sultan H. Almotiri ²

¹   Department of Computer Science, Garrison University, Lahore 94777, Pakistan; arfan137nagra@gmail.com (A.A.N.); iqramubarik914@gmail.com (I.M); mugeesasifm@gmail.com (M.M.A.)

²   Computer Science Department, Umm Al-Qura University, Makkah City 21961, Saudi Arabia; maeghamdi@uqu.edu.sa (M.A.G.); shmotiri@uqu.edu.sa (S.H.A.)

*              Correspondence: arfan137nagra@gmail.com (A.A.N.)       

 

 

First of all, let us express our most sincere thanks to Reviewers, who have provided a number of useful suggestions and assisted us to improve our work. Frankly speaking, if there were no constructive comments and suggestions, it was not possible for this paper to be improved in such a good shape with the updated a new version

.

The main revisions according to the Reviewers are made as follows:

 

Reply to Reviewer 3:

• For the quantitative analysis of the prediction models, the authors used only accuracy, precision, recall, and F1 scores. These metrics cannot give a comprehensive evaluation of the model. We would like the author(s) to add two metrics, G-mean and AUC, to increase the persuasiveness of the results.

 

Response: In the revised manuscript (applsci-1763996), comprehensive evaluation of the model with G-mean and AUC is performed to increase the persuasiveness of the results that have been added and calculated at the page no. 7 and 8.

                    

Table 3. The Comparative Analysis of G-mean and AUC with different models using the Haberman dataset.
Model	G-mean	AUC
FPSO-SVM + SMOTE [22]	0.6942	0.6813
PSO-SVM + SMOTE [22]	0.5832	0.6131
SVM+ SMOTE [24]	-	0.6096
KNN+ SMOTE [23]	0.6572	0.6649
Proposed Method	0.7897	0.6989

 

 

 

 

 

 

 

 

 

 

In Table 3, the technique FPSO-SVM + SMOTE has performed relatively better than the other two techniques, PSO-SVM +SMOTE, SVM + SMOTE and KNN + SMOTE. In our proposed method, the G-mean accuracy is 0.7897 and Area under the curve (AUC) is 0.6989, which is superior to all other techniques. It is evident not only from the accuracy levels but in terms of computational power that our technique has outperformed all the other state of the art algorithms.

 

• The authors used the dataset produced and published by Wroclaw Thoracic Surgery Center, which is identical to Jiang's article Hybrid Method to Predict Postoperative Survival of Lung Cancer Using Improved SMOTE and Adaptive SVM. We hope that the author(s) will add a comparison experiment to illustrate the superiority of this article over Jiang's method.

 

Response: The research work by Jiang in the paper titled as Hybrid Method to Predict Postoperative Survival of Lung Cancer Using Improved SMOTE and Adaptive SVM has been thoroughly compared not only by the proposed method but a few other relative works have also been considered. The following comparative tables have been added in the manuscript at the Page no. 8.

 

 

Table 2. Comparison of existing Approaches with respect to Accuracy and F1 score.
Model	Accuracy	AUC
FPSO-SVM + SMOTE [22]	0.8792	0.8807
PSO-SVM + SMOTE [22]	0.8713	0.7602
SVM + SMOTE [24]	0.7979	0.7966
KNN + SMOTE [23]	0.7708	0.7736
GA-SVM+ SMOTE Proposed method	0.90	0.78

 

 

 

    

 

 

 

 

 

 

 

 

Table 4. Accuracy and AUC comparison for different algorithms on the Haberman dataset.
Model	Accuracy	F1 score
RANDOM FOREST [1]	0.83	0.91
FPSO-SVM + SMOTE [22]	0.6890	0.6612
PSO-SVM +SMOTE [22]	0.6435	0.5089
SVM + SMOTE [24]	0.6291	0
KNN + SMOTE [23]	0.6630	0.6545
GA-SVM+ SMOTE Proposed method	0.85	0.91

 

 

 

 

 

 

Table 5. Accuracy and AUC comparison for different algorithms on the appendicitis dataset.
Model	Accuracy	AUC
FPSO-SVM + SMOTE [22]	0.6890	0.6813
PSO-SVM + SMOTE [22]	0.6435	0.6131
SVM + SMOTE [24]	0.6291	0.6096
KNN + SMOTE [23]	0.6630	0.6649
GA-SVM+ SMOTE Proposed method	0.85	0.69

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

As shown in Table 2, Random Forest has performed relatively superior than all the other four methods. It has achieved 83% classification accuracy, which is better than rest of the techniques. In Table 4, GA-SVM + SMOTE has outperformed the other techniques. Its Accuracy and AUC are 0.85 and 0.69 respectively, which are better than any of the other classifiers. In Table 5, the proposed method has performed better than the PSO + SMOTE, SVM + SMOTE, and KNN + SMOT algorithms.     

 

• The manuscript is very irregular in its use of acronyms, with several acronyms used directly without explanation. There are a number of acronyms where the explanation appears in the second or third use rather than the first use.

 

Response:  In the revised manuscript, all the acronyms have been defined according to the standard instructions.

 

• Chapter 3.1 uses the letter 'C' directly without any explanation, which makes the reader confused.

 

Response: The correction has been performed and authors are thankful for highlighting the mistake.

 

• Please check why the correct death prediction curve in Figure 3 exceeds the value 1 at class1, which is not reasonable. Also, please replace Figure 2 with Figure 3, their pixel values are too low making it difficult for the reader to read.”

 

Response: In the revised manuscript, all the figures have been updated according to the feedback. The death prediction curve in the Figure 3 has been redrawn and its limits are readjusted. The pixel values are enhanced with the standard resolutions in Figure 3.

 

The corresponding author on behalf of all the authors is thankful to the reviewer for providing valuable comments and feedback that has assisted in enhancing the quality of the manuscript.

 

 

 

 

Sincerely yours,

                        Dr. Arfan Ali Nagra,

 

 

 

 

 

 

Author Response File: Author Response.docx