Enhancing Channelized Feature Interpretability Using Deep Learning Predictive Modeling

Round 1

Reviewer 1 Report

The manuscript is inadequately documented with poor English and needs major revision in order to reach an acceptable level for scientific publishing. I think that some verb tenses in some sections should be revised (e.g., lines 309, 339, 417) and the use of certain articles as well.

 In addition, the figures need to be improved in general, for example in Figure 1, the text is in pink and needs to be replaced with an appropriate color.

Please add titles including (a), (b), etc for the figures instead of referring left, center and right. Each title should be explained in the caption and also inside the text and discussed the results.

Section 1 and Section 2 need to be rewritten completely and cited to the most representative references. All detailed information about the methods should be presented in Section 2. Therefore, please remove them in Section 3. I also have struggle to find any answers to my simple question, “The methods are appropriate to analyze the data.” Moreover, I am not sure the title of Section 2.2 is useful.

Lines 118-121: How are the labels chosen? The explanation is not satisfied for a scientific paper.

Line 273, Imbalance issue is known as a cross-validation in data science. Please give more information about this term.

Line179, the term, “nowadays” is not scientific. Please remove it.

Figure 4, there are no titles for the vertical and horizontal axis and also the legend is very wide with inadequate information. It is better to present area under curve (AUC) for each of the plots.

What are the green signs in Figure 6?

Lines 354-358, are these lines explaining the Figure 6? How did you quantify the accuracy in line 357? How is the ground truth reconstructed? Most important topological features of the ground truth should be presented.

Lines 413-415: How is possible to use some information from the predicted system to conclude the structural properties of the inputs? Otherwise, the expression, “predictions” is referred to different concepts in your study and it does not clarify for the readers.

Author Response

Dear Reviewer,

Kindly find my response, as attached.

Thank you.

Salbiah Mad Sahad

Author Response File: Author Response.docx

Reviewer 2 Report

Please remote the abbreviation in the title.  The introduction is very scanty 

1- There are no advanced machine learning models were developed. Those are existed model (CNN), however, probably not introduced for this engineering problem. Please compare with other stand-alone models but using one model will not justify the arguments

2- By looking at the abstract section, this section does not really present an abstract. You should follow those elements, i. problem statement or research motivation, ii. Main research aim, iii. Research results, general finding.

3- In the introduction section, you have explained the problem nicely, but you did not give credit for machine learning to solve this problem.? I see no problem statement clearly presented as well.

4- You better explain the data in section 2 in single section.

5- Methodology shall be reported in section 3. Even though no single theory for the CCN model

6- You have to define all presented variables of the model's formulations.

7- Cite proper references for the methodology.

8- All methodology diagrams are not well presented and visualized. You need to redraw figures in a way clear for readers.

9- The reference style is wrong, you need to revise this.

10- The discussion phase needs further debate and discussion.

11- There so many figures? Probably you need to merge some or move some them to supplementary.

12. The paper lacking discussion and comparison with other studies

13. The authors did not clearly state the direction of the paper because it is explaining both the classification and predictions dimension pls specify 

14. The conclusion is not in line with the context of the paper. There is no future research, neither limitation of the study 

Author Response

Dear Reviewer,

Kindly find my replies, as attached.

Thank You.

Salbiah Mad Sahad

Author Response File: Author Response.docx

Reviewer 3 Report

The article discusses a deep learning model for automating the predictions of geobodies using small amount of labelled training data. In general, the conducted research is worthy of attention and can be published in Applied Sciences Journal. However, there are some flaws/ambiguities in the present form of the article which need to be fixed in the revised manuscript. I have the following comments.

1.      Abstract appears as standalone entity for the entire article, it is not a good practice to use abbreviation in the abstract. Use the full word along with abbreviation for IU.

2.      The first three paragraphs of Introduction seem to specific to the topic and are difficult to understand for the general readership. I recommend adding a new paragraph showing the background of the work for general readership with relatively less background in the subject area.

3.      Use the reference number in the usual format of Sensors Journal for "Roden et al (2015)" in the caption of Table 1.

4.      It is not recommended to use abbreviations in the title of a topic. Use full name and its abbreviation for "SEG” in section 1.1

5.      In the section 2 on Methodology, the authors have directly mentioned the use of deep learning instead of machine learning without mentioning any specific reason. In the revised manuscript, clearly mention the reason for choosing deep learning over shallow machine learning.

6.      The statement for "High computational facilities are critically needed to train these models, together with numerous amounts of labelled datasets." is not correct for shallow machine learning and describe the implementation of deep learning. Clearly mention the word deep learning with it to convey the true meaning.

7.      Add a basic description of "PyLT" in section 2.2 and its working principle.

8.      The title and discussion of section 3.2 does not match. The title says "3.2 Deep Learning (DL) in Features Identification" however, there is no discussion about the feature identification and learning by deep learning model. Either change the title to be more descriptive or add discussion showing the feature identification capability of deep learning models.

9.      The word "multi-linear" in the statement "Conventional Machine Learning (ML) method were applied to build multi-linear perceptron (MLP)" on page 6 seems to be "multi-layer perceptron" rectify it in the revised manuscript.

10.  For the FCN employed in the current work, add a table showing full architectural details such as arrangement of layers, drop out probability of drop out layers, number, and size of filters in convolutional layers, etc. of the network.

11.  In section 3.4, the authors mentioned that they have used sixteen seismic attributes. I recommend to list those features with the statement or in the form of a table and their numerical counterpart for training the model.

Author Response

Hello,

Thanks for the comments. Attached are the replies for each question. Thank You.

Salbiah

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

I accept it 

Author Response

Dear Reviewer,

I've revised spell check/grammatical errors.

Reviewer 3 Report

The authors have addressed and incorporated my comments which improved the overall quality of the article. 

Author Response

Dear Reviewer,

Thank you and noted. Already revised spell checks/grammar checks.