An Intelligent Network Traffic Prediction Scheme Based on Ensemble Learning of Multi-Layer Perceptron in Complex Networks

Round 1

Reviewer 1 Report

Abstract: The actual findings and the objective of the research should be highlighted in the abstract.

Introduction: The contribution of the research is missing in this section and the limitations of the previous research also not discussed properly. Improve the discussion.

Related Works: GRU cell architecture is well known why again the author discussed the same? Instead include an architecture diagram. This section does not make any sense, better remove this portion and concentrate on the proposed model.

Methods: This section is too short and discussed methods are well known. What is the novelty here?. Similarly, the block diagram of the research is not included in this section. We recommend you to include block diagrams. Teacher forcing model is not discussed properly. Overall, the proposed section is short and weak, it's hard to understand the effectiveness of the proposed model. 

Input data/dataset information is missing and not discussed properly. Include the content.

Result and discussion: First of all the comparative analysis is missing. The author did many analyses based on quantitative models. We suggest you include qualitative and comparative analysis. Compare recent year papers to show the effectiveness of the proposed model. The units of performance analysis are improper in table 1 & 2, error (MSE,SMAPE,MAR) should be a number not a percentage. 

Figure 6-17: Most of the comparison is discussed between proposed and LSTM,GRU. It's totally unfair to compare the ensemble models in this way. Compare with other (minimum 5) deep learning models to show the effectiveness of the proposed model.

Conclusion: What is the actual findings and how its achieved is not discussed anywhere, we recommend the author to include the finding based discussion.

Reference: Papers 14,17,18,20,22,25,26,27,39,41 - These references are insufficient to support this research. Include standard journal references.

Improve the figure quality.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The work proposes a new scheme based on Machine Learning for processing data relating to traffic management. It turns out to be interesting from an engineering point of view and could also inspire further methodologies. Below are my suggestions and comments:

   1)     I suggest changing the introduction. In particular, it would be better to include more qualitative and review details of the problem in the introduction. Moreover, it would be better to strengthen section 2 where the background is reported. For example, the " problem of gradient disappearing in backpropagation" (on line 138) could be better explained. Again, show some details of the LSTM neural network and so on. In essence, deepen, in a slightly more quantitative way, what is reported in the introduction.

   2)     In the encoder stage (subsection 3.1), when you say that hi is the eigenvector of the i-th gate, while p is the number of eigenvectors on each gate, do you mean that p is the number of eigenvectors associated with the same port day to day? That is, are there p eigenvectors hi associated with the generic same port i?

   3)     Could you rephrase the following statement to make it clearer: "If the result is inconsistent with the expected result, there will be a problem: Whether to take the result as the next time's input or the correct result as the input of the next time" (lines 222-224)?

   4)     In relations (8)-(11) it should be specified more clearly what the parameter "m", associated with the summation with index j, indicates. I think "m" matches uppercase "M" (line 265).

   5)     It would be better to specify that the total number of samples that have been considered is 4000.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

I thank the authors very much for the work done.