July 11, 2024
Revision R2_3
===========
My initial perception of this document's contents was good and positive. The document is well-formatted, with text, figure plots, and "all the rest". However, we are now in 2024 and there is now available, for anyone to use, these technologies we all know on "social media" by "ChatGPT". So my assessment below takes into consideration all automation writing tools and apps I know to date.
Revision Files produced by this reviewer =============================================
The files and documents for the review made to this document can be found in the following data repository:
https://doi.org/10.7910/DVN/XCHMSE
What I did not analyze nor evaluate ============================
Text content, its quality, and grammar. I will leave this task to someone more specialized to this type of analysis and evaluation.
✓ corrected >> Figure 6 is confusing as it is. authors need to reposition it so it does not span 2 pages.
Experimental Data ====================================
Data availability
✓ A research document with no data is ... a blog post or an internet article. Authors need to add a data source, for instance, a repository that a reviewer can use when performing analysis and evaluating the document.
The data availability statement is now present, and the link to a repository WAS MADE AVAILABLE TO THIS REVIEWER.
https://github.com/Awu-SEU/EMI-Data
Artificial Neural Networks Modeling ========================
It is missing, in the document, the following :
✓ - is missing a link to a repository with the Python code programmed for this work.
✓ - The algorithms and customizations made to the models are missing in the document. What was the loss function? Optimizers used? Is any customization made to the models?
The authors added an "objective function", used to do model evaluation.
✓ an explanation is provided in this revision - Figure 11 is ...dubious and it can be misleading. It requires a more elaborate and detailed explanation
✓ added a biblio. ref. - The authors need to justify clearly why they opted to divide the train and test datasets 80/20.
✓ added to section 5.1 - normalization of experimental values is not properly justified. This is an analysis technique, broadly used, with known drawbacks that need to be presented and discussed in this document.
✓ - Nowadays with the availability of modeling automation tools, it has become a simple and easy task to plot "some charts" and a table with almost perfect accuracies near 100%. Without the availability of experimental data and the algorithms and programming code, it is fair for this reviewer to question experimental data origins and how modeling was made.
The data availability statement is now present, AND THE LINK to a repository WAS MADE AVAILABLE TO THIS REVIEWER.
Summary ================================================================
A work like the one presented here, using accuracy and typical statistical mean values ( RSME, MAE, MAPE) is insufficient to characterize the performance of the models discussed. It is well-known that "accuracy" is a good performance indicator in global analysis and assessment. This work shows it too, with accuracies nearing 100%, however, this is not enough. It required more, other statistical parameters, preferably ones that are able to infer performance with more detail and less "globally" / superficial. If published like this, it will have much less scientific value to the authors and to the Journal.
R1 ------------------------------------------------
see previous response. thank you
R2 ----------------------------------------
✓ In this document revision, R2, the authors removed the link to a data repository. A justification for why they choose to do that is needed.
The XAI section is welcoming to this reviewer, but not as it is presented, for the same reasons presented before, which can be summarized as follows: "A research work without data is a.... blog post. "
The data availability statement is present, AND THE LINK to a repository WAS MADE AVAILABLE TO THIS REVIEWER.
R2_2 ----------------------------------------------
✓ In this document revision, R2_2, the authors added again the link to a data repository for reviewers to use. however,
THE LINK IS NOT AVAILABLE ON THE DOCUMENT ITSELF.
Although this is not a mandatory requirement this reviewer strongly advises it to add it to the document. Even when equating intellectual property rights and protection (patent).
✓ The text throughout the document focuses on the importance and influence of Temperature to model with accuracy. With the inclusion of section "5.3 Shapley Additive Explanations (SHAP)" in the document revision R2, the authors state:
"the information characteristics of the impedance (such as frequency, temperature value, and peak) have little impact on the model prediction results"
AND
"This shows that although the temperature has a great influence on the impedance, in the prediction process, the model is mainly based on the prediction of the subinterval area of the signal segmentation"
A more elaborate explanation is provided in the revision of the document.
❌ Moreover, the last paragraph in section 5.3 is poor in discussing signal labels and their relation to prestress.
R2_3 ----------------------------------------------
❌ Figure 5 - The label "wireless piezoelectric sensor" for the Data Acquisition Devices is not correct
This reviewer strongly advises authors to make available in a data repository the source code of the software program programmed in QT 5.8 and include a link to it in the research document. I will add immense value to the author's research work.
❌ In the abstract...
- the word "temperature" is repeated too many times.
- consider reducing the number of words in the abstract
From the Spearman correlation coefficient diagram in Figure 11, it is clear that temperature strongly correlates with "Area 2" which is not properly identified by the ML Models investigated.
From the "Feature importance analysis utilizing the SHAP" in Figure 14 it is clear "Area 2" has an "abnormal" weight on the XGBoost Model while at the same time having a strong correlation with temperature (Fig. 11). This requires further investigation and analysis IF the author's wish to improve quality of their finding and work presented.
In summary,
the document requires minor revision to address the main points stated above in section R2_3.
Final Recommendation ===============================================================
My final recommendation is "Amendments required before acceptance" requiring minor revision.
For this reviewer to accept for publication, what was previously stated needs to be present in all revisions of the document until final acceptance.
one side note ==================================================================
If the author's main concern is about "intellectual property", this reviewer strongly advises them to first take care of such "bureaucratic" tasks, followed by the publication of both experimental data and the algorithms for each model in a data repository, and only after all that is completed and made available, submit this document to the journal to be evaluated for publication.
On a final note to both Editors and Authors ===========================
I find necessary, the inclusion of a statement by the authors about the usage of any assisted writing tools. From simple MS Word add-ons such as Grammarly and Writefull to test automation tools such as ChatGPT and even more elaborate algorithms, able to deliver a complete document with text and graphics ready to be edited and replaced with simple copy-and-paste functionality.