Real-Time, Model-Agnostic and User-Driven Counterfactual Explanations Using Autoencoders

Round 1

Reviewer 1 Report

1.      In current version, the title reflects that the authors performed centrifuge model tests and they discussed the test result. The title must reflect the study novelty.

2.      Revise/ rephrase the structured of abstracts, and must follow/include but without headings: 1) Background 2) research problem 3) materials and Methods 4) Results and discussion and 5) Conclusion.

3.      The authors have discussed their test results, however, the authors have not compared their results with the literature.

4.      Most of the literature is related to the field problem due to heavy rain. However, the centrifuge model test studies for such tests has not been referred in the study.

5.      The final part (conclusion part) must be concise and be based on the model test results.

6.      The figures and table caption must be enough for self-explanation.

Author Response

Expressing gratitude for the recommendations provided, please refer to the enclosed document for detailed responses to the raised concerns.

 

Author Response File: Author Response.pdf

Reviewer 2 Report

The proposed method RTGECx is well explained. I have the following questions.

1. What is the need to replicate input data?

2. How does this technique differ from conventional ANN or CNN? 

3. Detailed interpretation of results in Table5 is needed for clearer understanding of the proposed technique.

4. In both figures 10 and 11, the original and counterfactual signals are exactly same. How can u attribute this to the proposed method?

5. Why is MSE and other statistical performance measure parameters not found as like Table 5 for all methods?

Author Response

Expressing gratitude for the recommendations provided, please refer to the enclosed document for detailed responses to the raised concerns.

 

Author Response File: Author Response.pdf

Reviewer 3 Report

The manuscript is well organized and well written. Thus, I accept it for publication. 

Author Response

Expressing gratitude for the recommendations provided, please refer to the enclosed document for detailed responses to the raised concerns.

Author Response File: Author Response.pdf

Reviewer 4 Report

The paper entitled " Real-time model agnostic counterfactual explanations" is very well written and structured, contributes to the field of counterfactual explanations. Authors propose a new method for generating user-driven counterfactual explanations in real-time using autoencoders. The autoencoder is trained with a multi-objective loss function to ensure that the generated counterfactuals remain within the data distribution, are valid counterfactuals, and align with the desired outcome. This method outperforms classical methods in terms of speed, efficiency, and interpretability of the generated counterfactuals. The method was evaluated on the MNIST handwritten digit dataset and a gearbox dataset, showing improved performance over other methods and low values in all terms of the loss function. The methodology is well described and the results have been well discussed. So, I recommend to be considered for publication. But I have however some remark about this paper:

·       - Review the first line of page 14, (function of the equation ??)

·       - It would be better to present the architectures of the models used in figures (Black-box model and Autoencoder used for MNIST)

·       - Rename section 5 to Results and Discussion instead of Result, and incorporate part of the conclusion into this section

Author Response

Expressing gratitude for the recommendations provided, please refer to the enclosed document for detailed responses to the raised concerns.

Author Response File: Author Response.pdf