SCConv-Denoising Diffusion Probabilistic Model Anomaly Detection Based on TimesNet

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This research is interesting and aligns well with the scope of the Electronics journal. I suggest a few revision points below, and I hope these comments will be helpful in enhancing the quality of the manuscript.

1) Kindly emphasize the core contributions of the study more clearly by providing a more detailed explanation of the innovative aspects of integrating SCConv and DDPM.

2) In addition to the evaluation metrics used (F1, P, R), I suggest including additional performance indicators such as AUC scores, possibly with figures, to provide deeper insights.

3) Kindly provide detailed information on hyperparameter settings, model implementation code, and data preprocessing methods to improve the reproducibility of this research.

4) I suggest the authors elaborate on the limitations mentioned in the conclusion (e.g., resource consumption issues when handling large datasets) and propose concrete research directions to address these limitations.

Author Response

Comments 1 : Kindly emphasize the core contributions of the study more clearly by providing a more detailed explanation of the innovative aspects of integrating SCConv and DDPM. 

Response 1 : Thank you for pointing this out. We agree with this comment. Therefore, we have emphasized the core contributions of the study more clearly. This change can be found in page 2,line 62-73

Comments 2 : In addition to the evaluation metrics used (F1, P, R), I suggest including additional performance indicators such as AUC scores, possibly with figures, to provide deeper insights.

Response 2 : Thank you for pointing this out. Unfortunately, it is difficult to find papers comparing AUC experimental indicators in the field of time series anomaly detection, and the AUC calculated in this method does not demonstrate outstanding performance compared to other models. But in order to prove that our method is progressiveness, we added a new public dataset to compare with other benchmark models, and achieved relatively good experimental results. This change can be found - page 12 , table 3.

Comments 3 : Kindly provide detailed information on hyperparameter settings, model implementation code, and data preprocessing methods to improve the reproducibility of this research.

Response 3 : Agree. We have, accordingly, revised  hyperparameter settings and  data preprocessing methods to emphasize this point. This change can be found - page 11, section 4.4 and table 2

Comments 4 : I suggest the authors elaborate on the limitations mentioned in the conclusion (e.g., resource consumption issues when handling large datasets) and propose concrete research directions to address these limitations.

Response 4 : Thank you for pointing this out. We agree with this comment. We will add this part explanation in section Conclution. This change can be found -page 14, line 480 - 483

Reviewer 2 Report

Comments and Suggestions for Authors

This article introduces a promising method for anomaly detection in time series data by integrating TimesNet, Denoising Diffusion Probabilistic Models, and the Spatial and Channel Reconstruction Convolution module. This method transforms 1D time series into 2D tensors, enabling advanced feature learning and robust anomaly detection. Experimental results on real-world sensor datasets demonstrate superior performance compared to existing methods in terms of precision, recall, and F1-score.

Below are my suggestions for improving the article:

1. I recommend changing the section title from "Methodology" to "Materials and Methods." In the Methodology section of an article, the aim is to explain how the research was conducted.

2. What specific Inception model architecture is used for capturing 2D temporal variations?

3. Regarding SCConv, can you provide more details on the hyperparameters used, such as the number of groups (g) for Group-wise Convolution?

4. The description of the experimental setup is somewhat vague. Clarifying the sliding window size used for data segmentation would be beneficial. The statement "we employ the classic reconstruction error as the common anomaly detection criterion for all experiments" needs further explanation. How is this reconstruction error calculated for each baseline model?

5. I suggest including a list of abbreviations.

Author Response

Comments 1 : I recommend changing the section title from "Methodology" to "Materials and Methods." In the Methodology section of an article, the aim is to explain how the research was conducted.

Response 1 :  Thank you for pointing this out. We agree with this comment. Therefore, we have changed the section title from "Methodology" to "Materials and Methods." This change can be found – page 4.

Comments 2 : What specific Inception model architecture is used for capturing 2D temporal variations?

Response 2 : Thank you for pointing this out. The Inception module we utilized in our paper was originally proposed by Christian Szegedy et al. in their seminal work titled "Going Deeper with Convolutions", which introduced the GoogleNet architecture.

Comments 3 : Regarding SCConv, can you provide more details on the hyperparameters used, such as the number of groups (g) for Group-wise Convolution?

Response 3 :  Agree. We will add the number of groups (g) for group-wise convolution into the hyperparameter list. This change can be found – page number, paragraph, and line. This change can be found - page 11 , table 2

Comments 4 : The description of the experimental setup is somewhat vague. Clarifying the sliding window size used for data segmentation would be beneficial. The statement "we employ the classic reconstruction error as the common anomaly detection criterion for all experiments" needs further explanation. How is this reconstruction error calculated for each baseline model?

Response 4 : Thank you for pointing this out. In response to the comment, "Clarifying the sliding window size used for data segmentation would be beneficial," we will incorporate an analysis of the sliding window parameter into the hyperparameter sensitivity study. This addition will systematically evaluate the impact of varying window sizes on data segmentation and model performance, ensuring methodological transparency and robustness. We thank the reviewer for highlighting the need for clarity regarding reconstruction error calculation. As noted in Section 3.4 , the Mean Squared Error (MSE) is employed as the standardized metric to compute reconstruction errors for all baseline models. We will supplement this point immediately following the mentioned sentence. This change can be found -page 11,line 398

Comments 5 : 

Response 5 : We agree with this comment. We will include a list of abbreviations in the manuscript to enhance clarity and readability. This change can be found -page 15,line 491

Reviewer 3 Report

Comments and Suggestions for Authors

The authors have presented a paper titled "SCConv-Denoising Diffusion Probabilistic Model Anomaly Detection Based on TimesNet". The authors are suggested to clarify the following comments.

1. The abstract should include the performance of the SCConv approach.

2. The introduction part should mention the novelty of the work.

3. Experiment results should include more benchmark dataset comparison.

4. Figure 5. should include a more clear picture( Framework of the SCConv)

Comments on the Quality of English Language

The Quality of the English Language can be improved.

Author Response

Comments 1 : The abstract should include the performance of the SCConv approach.

Response 1 : Thank you for pointing this out. We agree with this comment. Therefore, we will add this part of content to the abstract. This change can be found in page 1.

Comments 2 : The introduction part should mention the novelty of the work.

Response 2 : Thank you for pointing this out. We agree with this comment. Therefore, we will add this part of content to the abstract. We will explain the novelty of this method in the introduction section. This change can be found in page 2.

Comments 3 :  Experiment results should include more benchmark dataset comparison. 

Response 3 : Thank you for pointing this out. We agree with this comment. Therefore, we will add dataset SWAP in experiment section. This change can be found in page 12, table 3.

Comments 4 : Figure 5. should include a more clear picture( Framework of the SCConv)

Response 4 : Thank you for pointing this out. We agree with this comment. Therefore, we will change more clear picture.This change can be found in page 8.

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The quality of the manuscript has significantly improved.