Multivariate Temporal Convolutional Network: A Deep Neural Networks Approach for Multivariate Time Series Forecasting
Round 1
Reviewer 1 Report
I congratulate you because this article is very interesting and with a pleasant and simple reading to read.
These are some of the corrections that the article should have:
This article introduce a novel M-TCN model and compared with traditional models and rich NN based competitive models.
On page 6 dataset is NOT available online: https://archive.ics.uci.edu/ml/datasets.html
Please, review the URL.
Table 4. Results summary (in RMSE, RSE and CORR) of all methods with two datasets:needs a better explanation in natural language.
Explain better figures 6 and 7 in order to distinguish the two differences.
Conclusions need to be based more on the results given in the experiments that were elaborated with both datasets.
The bibliography is poor, it is necessary to deepen the state of the art.
Author Response
Dear Reviewers:
We took considerable modifications based on all your nice comments. The answer to your points are listed below.
-------------------------------------
On page 6 dataset is NOT available online: https://archive.ics.uci.edu/ml/datasets.html
>>>>>>>>>>>
Thank you. The link has been updated.
-------------------------------------
Table 4. Results summary (in RMSE, RSE and CORR) of all methods with two datasets:needs a better explanation in natural language.
>>>>>>>>>>>
Thank you. More about detailed algorithms of M-TCN has been implemented. And the description of this table has been updated, which is more clean and readable.
-------------------------------------
Explain better figures 6 and 7 in order to distinguish the two differences.
>>>>>>>>>>>
Thank you. Figure 6 and Figure 7 are for PM2.5 and ISO-NE in seperately. The amplitude spectrum analysis is updated to be more detailed.
-------------------------------------
Conclusions need to be based more on the results given in the experiments that were elaborated with both datasets.
>>>>>>>>>>>
Thank you. The conclusions are updated.
-------------------------------------
The bibliography is poor, it is necessary to deepen the state of the art
>>>>>>>>>>>
Thank you. More references are implemented.
Reviewer 2 Report
The core deep learning algorithm, M-TCN is poorly presented. Presenting the block diagrams (Figs 3 and 4) without following a detailed sound mathematical representation and explanation is not adequate.
In page 6 the following is stated: "The residual block 2 is shown in Figure 3 (right), which has the same basic structure, but one unit layer is implemented." --> What does this mean?
Overall, the description of the M-TCN algorithm is poorly presented. It requires a significant improvement.
Author Response
Dear Reviewers:
We took considerable modifications based on all your nice comments. The answer to your points are listed below.
----------------------
The core deep learning algorithm, M-TCN is poorly presented. Presenting the block diagrams (Figs 3 and 4) without following a detailed sound mathematical representation and explanation is not adequate.
>>>>>>
Thank your. More about detailed algorithms of M-TCN has been implemented.
----------------------
In page 6 the following is stated: "The residual block 2 is shown in Figure 3 (right), which has the same basic structure, but one unit layer is implemented." --> What does this mean?
>>>>>>
Sorry it makes confused. It changes in this version.
----------------------
Overall, the description of the M-TCN algorithm is poorly presented. It requires a significant improvement.
>>>>>>
Thank you for your significant this comment. More about detailed algorithms of M-TCN has been implemented.
Reviewer 3 Report
The authors propose a multivaritate temporal approach based on CNN to improve the prediction of multivariate time series forecasting.
The proposed approach is really interesting but there are some points that the authors may better explain. Some experiments on efficiency of the approach should be added. Furthermore, the authors should provide more details about the used technologies.
The related work should be extended with other more recent works that may leverage also multimedia contents. See for instance:
1) Multimedia summarization using social media content. Multimedia Tools and Applications, 77(14), 17803-17827.
Finally a linguistic revision is necessary.
.
Author Response
Dear Reviewers:
We took considerable modifications based on all your nice comments. The answer to your points are listed below.
-------------------
The proposed approach is really interesting but there are some points that the authors may better explain. Some experiments on efficiency of the approach should be added. Furthermore, the authors should provide more details about the used technologies.
>>>>>>>>>>>>>
Thank you for your nice comment. More about detailed algorithms of M-TCN has been implemented in mathematical and descriptions. Besides, the results on efficiency are also presented in this version.
-------------------
The related work should be extended with other more recent works that may leverage also multimedia contents. See for instance:
1) Multimedia summarization using social media content. Multimedia Tools and Applications, 77(14), 17803-17827.
>>>>>>>>>>>>>
Thank you. It's great. We have cite this article, as well as other related works recently.
-------------------
Finally a linguistic revision is necessary.
>>>>>>>>>>>>>
Thank you. The paper has been polished. Hope it is more readable and scientific.
Round 2
Reviewer 2 Report
The improvement of the paper from the original is satisfactory. No other comments.
