Early Earthquake Detection Using Batch Normalization Graph Convolutional Neural Network (BNGCNN)

Round 1

Reviewer 1 Report

Earthquake is a major hazard to humans, buildings, and infrastructure. Early automatic detection of an earthquake from raw waveform data using deep learning models like graph neural network (GNN) is becoming an important research area. Multi-layered structure and too many epochs require plenty of time for training these models. It is also hard to train the model with saturating nonlinearities. The batch normalization technique is applied to each mini-batch to reduce epochs in training and obtain a steady distribution of activation values. It improves model training and prediction accuracy. This study proposes a deep learning model batch normalization graph convolutional neural network (BNGCNN) for early earthquake prediction.

1. Some figures are with low resolution.

2. No overview of the proposed method.

3. No comparison with other similar algorithms.

4. No merits to evaluate the performance.

Author Response

We are thankful to all the reviewers and editors for their valuable feedback that help and guide us to improve the quality of our manuscript. We have tried our best to fix all the identified gaps, and technical issues raised by the honorable reviewers and the editor. To understand the changes in the manuscript, we have highlighted the text in the manuscript.

Author Response File: Author Response.pdf

Reviewer 2 Report

The article presents a machine learning research and its results for early earthquake prediction.

 

The achievements of the research are promising.

The quality of the documents has to be enhanced. There are some formatting, grammatical and logical mistakes. See some detailed examples below.

 

lines 19-20: Multi-layered structure and too many epochs require plenty of time for training these models.

The sentence is not appropriate in a scientific journal article.

 

lines 29-30: our method shows promising results and superior performance than the baseline model GNN

bad grammar

 

line 36: Earthquake is the major hazard

the -> a

 

lines 43-44: different font style/size

 

line 51: A problem with deep learning models is hyper-parameter tuning.

No, hyper-parameter tuning is not a problem.

 

lines 57-59:

check font for "prediction"

check sentence

 

line 166: it directly operates -> it operates directly

 

(4): TanH(x)  ..... ∊ [−1,1]

Should be (-1, 1)

There is an error in the formula as well.

 

line 225: We used Adam optimizer with learning rate of 0.0001 to train the model.

I guess it was not constant learning rate, rather you mentioned the starting learning rate.

 

Figures 8-9: hard to see figures

 

Figure 10

Unbelievable results, hard to see, but train and validation lines seem to be very close. The "No station location" case is not clear, what does train + validation mean?

 

Not all the problems are mentioned above, please provide a careful revision.

 

Author Response

We are thankful to all the reviewers and editors for their valuable feedback that help and guide us to improve the quality of our manuscript. We have tried our best to fix all the identified gaps, and technical issues raised by the honorable reviewers and the editor. To understand the changes in the manuscript, we have highlighted the text in the manuscript.

Author Response File: Author Response.pdf

Reviewer 3 Report

The manuscript entitled “Early Earthquake Prediction using Batch Normalization Graph 2 Convolutional Neural Network” investigated a new model  for early and rapid system to predict possible earthquake.

After careful review, the manuscript has a reasonable effort and technical information. Firstly, I am not sure about the validation procedure and how is the applicability and acceptance of the method confirmed and compared. It has to be clarified in the next response letter. However, there is not much novelty on the work as there are many similar works, and there are some points that must be considered in the revision to be worth being accepted. Therefore, I strongly recommend the authors to follow the comments below:

1- Abstract does not present important points. It should be between 250 to 300 words and concisely mention the problems of previous works and novelties in this paper.  It needs intensive paraphrasing and has to be written more scientific and prevent short sentences.

2-Similarly, the introduction has very poor structure and lack of literature review. Usually in the chapter of introduction the background and needs of this study and why it has to be highlighted and prepare readers to go further. Then your second chapter should be literature review where you present an overview on the previous works and the main problem statements of work and how it can be improved or overcome on it. You can concise it and point the important parts and highlights the achievements.

3- The work presents a poor literature review on the methods based on modern techniqes such as soft computing (ML, Fuzzy and AI) for earthquake prediction, seismic vulnerability and fragility/damage assessment. Because your work is not much new and novel, I strongly recommend you make your introduction and study on previous methods interesting for readers by adding the following new works which I found it new and related to your work which are based on different methods for earthquake engineering and related stuff. It will increase the depth of your review and stronger and wider area.

-A review on application of soft computing techniques for the rapid visual safety evaluation and damage classification of existing buildings

-Multi-Hazard and Spatial Transferability of a CNN for Automated Building Damage Assessment

-Earthquake hazard safety assessment of existing buildings using optimized multi-layer perceptron neural network

-DeepShake: Shaking Intensity Prediction Using Deep Spatiotemporal RNNs for Earthquake Early Warning

-Improved rapid assessment of earthquake hazard safety of structures via artificial neural networks

-On-Site Alert-Level Earthquake Early Warning Using Machine-Learning-Based Prediction Equations

-Earthquake vulnerability of city regions based on building typology: rapid assessment survey

There are also many other methods based on MLP and Fuzzy type-2 that can be found on the internet to add.

4- All Figures : Please improve the quality and font size to be more visible. Please make the graphic nice and well cropped. Size of figures, fonts and legend must have harmony, please revise it well. 

5- Please provide a figure that shows the architecture of your model and network. Something like a graphical abstract. Somehow you have presented in Fig. 1 but still needs to provide more details of your proposed work.

6- Please make your tables and figures follow a same path and font size and colors and adjust them in a proper way. Figures 8, 9 and 10 are very difficult to read and understand.

7- There are many typos that needs to be corrected and please make the format similar for all. Also, be careful about spaces after (.) amd (,) which in some cases are doubled or missed.

8- Please type all the equations, some of them seems to be a figure instead of typing. Some Figures are also seems to be like screenshot.

9- You did not highlight the problem statement, objectives and novelty of your proposed method; That is why increasing the background of literature review based on the recommended works can help in this manner.

10- There is a need for intensive proofreading the paper. The used language is so basic and does not sound scientific and professional academic work.

11- Please provide more information about the figures and tables and write more in the body of the text about them and the information they provide.

At the end as I have mentioned, there are not much significant novelty on this work but the efforts were good and it would be good if you revise it according to the points provided and other reviewers.

Author Response

We are thankful to all the reviewers and editors for their valuable feedback that help and guide us to improve the quality of our manuscript. We have tried our best to fix all the identified gaps, and technical issues raised by the honorable reviewers and the editor. To understand the changes in the manuscript, we have highlighted the text in the manuscript.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

The authors extensively corrected the manuscript, the quality of the paper is much better now.

There is, however, one issue for which I cannot yet accept the authors' answer.

 

For Figure 10 (Figure 11 in V2) my comment was:

Unbelievable results, hard to see, but train and validation lines seem to be very close. The "No station location" case is not clear, what does train + validation mean?

 

The authors' answer did not explain how can the results achieved on the validation data be so much close to the ones on train data. They are not only close but almost the same. From my experience this happens usually when the training and the validation data are not for well separated samples.

It is also not explained why is the plot for the case when no station location is available was done with combining the training and the validation data. Without distinct plots for the training and validation data we have less information about fittness/overfitting.

Further examining the figure it the plot (b)[d] seems very interesting. Is there any explanation for this strange curve?

 

Author Response

Dear Reviewer, We thank you for your valuable suggestions to improve the quality of our manuscript. We tried our best to solve all the issues pointed out by you. 

Author Response File: Author Response.pdf

Reviewer 3 Report

Dear Authors

Many thanks for your response and significant changes.

Author Response

Dear Reviewer, We thank you for your valuable suggestions to improve the quality of our manuscript. We tried our best to solve all the issues pointed out by you. 

Author Response File: Author Response.docx