Self-Organized Fuzzy Neural Network Nonlinear System Modeling Method Based on Clustering Algorithm
Round 1
Reviewer 1 Report
Comments:
The manuscript presents application of " Self-Organized Fuzzy Neural Network Nonlinear System Mod-2 eling Method Based on Clustering Algorithm", which is interesting. The subject addressed is within the scope of the journal. However, the manuscript, in its present form, contains several weaknesses. Appropriate revisions to the following points should be undertaken in order to justify recommendation for publication.
1. Topic/area of research
The topic seems important in the area of Fuzzy Neural Network. I would say that the author put reasonable effort in this direction. However, the study seems too general not very specific. There are many studies done in this particular area. Please clarifying the uniqueness of the study and add in text.
2. Introduction and review of literature
Comprehensive review of literature was given on the variables. However, different perspective studies, methodology or contradictory results of studies on the variables were missing. Studies on different perspective on the variables could be added in the text for reducing biasness.
3. Full names should be shown for all abbreviations in their first occurrence in texts.
4. The novelty and utility of the work should be highlighted in the introduction as well as discussion sections.
5. Materials and methods
The authors should use more robust model evaluation measure. I can’t see any measure used for investigating the Relationship of Plant Species and Sampling Area under Different Grazing Intensities. Number the equations in this section.
The following literatures could be helpful in understanding the different evaluation measures:
Gholami, et al. (2021). Comparison of self-organizing map, artificial neural network, and adaptive neuro-fuzzy inference system methods in simulating groundwater quality: geospatial artificial intelligence. Water Resources Management,
Gholami, et al. (2022). Estimation of soil splash erosion using fuzzy network-CANFIS. Arabian Journal of Geosciences,
6. I found the discussion section weak. It should be strengthened by comparing the results with the findings of the other studies.
7. The future scope and limitations of the work should be highlighted in the conclusion section.8. Figs have Medium quality.
Author Response
Dear Reviewer,
Thank you for allowing a resubmission of our manuscript, with an opportunity to address your comments.
We have revised some of the language in the article. And we are uploading (a) our point-by-point response to the comments, (b) an updated manuscript with yellow highlighting indicating changes.
Best regards,
Tong Zhang and Zhendong Wang
Author Response File: 
Author Response.docx
Reviewer 2 Report
The paper discusses nonlinear system identification using a type of SOFNN.
1. You are using (quasi) second order method for optimization while SOFNN uses simpler adaptive gradient method. It is expected that your result to be better with more computation cost while this is in reverse as Table 1 and 2 indicate. Why?
2. Have a comparison with the results of the newly published paper” An efficient self-organizing deep fuzzy neural network for nonlinear system modeling
Author Response
Dear Reviewer,
Thank you for allowing a resubmission of our manuscript, with an opportunity to address your comments.
We have revised some of the language in the article. And we are uploading (a) our point-by-point response to the comments, (b) an updated manuscript with yellow highlighting indicating changes.
Best regards,
Tong Zhang and Zhendong Wang
Author Response File: Author Response.docx
Reviewer 3 Report
An appealing on-line neuro-fuzzy model identification approach based on back-propagation error based update of the model parameters (centers, sigmas and consequent parameters). The approach is also able to add new neuron on the fly with new incoming data samples and to prune and merge older neurons not needed any longer. Due to the nice results, showing the performance capabilities of the new method, i recommend acceptance of the paper, subject to some issue for better clarification.
1.) As the approach is an evolving approach by being capable to add new neurons on demand, the authors should include a discussion and references to past evolving neuro-fuzzy approaches (ENFS), ideally to recent survey papers provide a comprehensive overview, such as:
Evolving Fuzzy and Neuro-Fuzzy Systems: Fundamentals, Stability, Explainability, Useability, and Applications, in: Handbook on Computer Learning and Intelligence, World Scientific, pp. 133--234, New York, 2022
Evolving Fuzzy and Neuro-Fuzzy Approaches in Clustering, Regression, Identification, and Classification: A Survey, Information Sciences, vol. 490, pp. 344--368, 2019
In this context, they authors should make better clear in which ways their approach improved current SoA techniuqes in ENFS.
2.) Regarding weight parameters adaptation in current ENFS approaches, please refer to the following paper, discussing possibilities with pros and cons:
Improving the Robustness of Recursive Consequent Parameters Learning in Evolving Neuro-Fuzzy Systems, Information Sciences, vol. 545, pp. 555--574,
2021
3.) you mention "the model takes the expert experience into account..." - where is that the case? - latter i see only pure data-driven learning aspects in your approach, so either please clarify or delete this sentence...
4.) The neuron evolution criterion in (15) include the the Mahalanobis distance of the new sample to already existing neurons - this idea is not new as presented in:
Generalized Smart Evolving Fuzzy Systems, Evolving Systems, vol. 6 (4), pp. 269--292, 2015
please include reference and discuss difference (guess the error criterion is additional in yours...)
5.) Activation strenght threshold equal to 0.2 - how is this motivated?
6.) Your neuron merging approach based on Gaussian cutting points is also not new as already presented in
On-line Elimination of Local Redundancies in Evolving Fuzzy Systems, Evolving Systems, vol. 2 (3), pp. 165--187, 2011
please include reference and discuss different aspects of yours compared to this approach...
7.) Equation (38) for updating all your model parameters requires the inverse of the matrix - how can this be quickly calculated?
8.) The derivations of (in)equations (55) and (56) need more explanations
Author Response
Dear Reviewer,
Thank you for allowing a resubmission of our manuscript, with an opportunity to address your comments.
We have revised some of the language in the article. And we are uploading (a) our point-by-point response to the comments, (b) an updated manuscript with yellow highlighting indicating changes.
Best regards,
Tong Zhang and Zhendong Wang
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
Thank you for your revised submission.
