LST-GCN: Long Short-Term Memory Embedded Graph Convolution Network for Traffic Flow Forecasting

Round 1

Reviewer 1 Report

Referee's Report

Title: LST-GCN: Long Short-term Memory Embedded Graph Convolution Network 2 for Traffic Flow Forecasting

Authors: Xu Han and Shicai Gong

MS. Ref. No. electronics-1797380

 

    In this study, the authors propose a model named LST-GCN to improve the accuracy of current traffic flow predictions. The authors simulate the spatiotemporal correlations present in traffic flow prediction by optimizing GCN (Graph Convolutional Network) parameters using an LSTM (Long Short-term Memory) network. Specifically, the authors capture spatial correlations by learning topology through GCN networks and temporal correlations by embed ding LSTM networks into the training process of GCN networks. The method improves the traditional method of combination recurrent neural network and graph neural network in the original spatiotemporal traffic flow prediction, so it can better capture the spatiotemporal features existing in the traffic flow. Extensive experiments conducted on PEMS dataset illustrate the effectiveness and outperformance of our method compared with other state-of-the-art methods.

    I have carefully gone through the manuscript and have some observations/remarks regarding it to improve its presentation.

 

1.Please remove the sentences "Traffic flow prediction is an important part of the intelligent transportation system. Accurate traffic flow prediction is of great significance for strengthening urban management and facilitating people's travel." from the abstract.

2.Line 24, "significance[1-2]." should be "significance [1-2].".

3.Line 41, "Methods[3]" should be "Methods [3]".

4.Line 43, "...information of the data, and have..." should be ""...information of the data and have...".

5.Line 52, "this paper, and the LSTM model[4]" should be "this paper and the LSTM model [4]".

6.Line 53, "model[5], to" should be "model [5], to".

7.The above remarks hold in for page 2, 3 and 4 before the section "3. Preliminaries".

8.Definition 1 and Definition 2 should be properly re-written.

9.Line 175, "," is missing and "and" is missing. 

10.Line 177, "." is missing.

11.Lines 193, 194, 195, 196, "," is missing and "and" is missing between lines 197 and 198.

12.Line 200, "c_{t-1}, and" should be "c_{t-1} and".

13.Line 222, "...Y_{d}, and Y_{w}, respectively." should be "...Y_{d} and Y_{w} respectively.".

14.Line 234, "together, and the representation" should be "together and the representation".

15.Line 242, "." is missing.

16.Line 245 and line 248, "." is missing.

17.Line 269, "set, and the" should be "set and the".

18.Line 277, "frequency of 2.7GHz, and a NVIDIA" should be "frequency of 2.7GHz and a NVIDIA".

19.Line 281, "200, the training batch size is 32, and the" should be "200, the training batch size is 32 and the".

 

    Such problems are throughout the manuscript. All the problems should be corrected.

    As far as the experimental results are concerned, these are fine. The graphical and tabular explanations are fine.

    The results are novel but the way in which they are presented should be significantly improved. Therefore I recommend the publication of the manuscript to "electronics" but after addressing the above suggestions and remarks.

Author Response

Thank you for your corrections, which we have made in the latest submitted manuscript.

Author Response File: Author Response.pdf

Reviewer 2 Report

The manuscript entitled “ LST-GCN: Long Short-term Memory Embedded Graph Convolution Network for Traffic Flow Forecasting” discusses the mothod for traffic flow prediction. To this reviewer’s opinion, the manuscript seems interesting and worthy of attention. However, there are some aspects that should be addressed or clarified before further consideration. 

1) The state of the art is well addressed but the novelty of this new method and its comparison with the previous ones should be highlighted in a more suitable manner. I kindly request the Authors to be referred to the novelty of the paper.  Espetially it should be clearly written  what is the contribution of the Authors in relation to other articles

2) The organization of the paper should be included at the end of the Introduction section.

3) The Authors should discuss the data set and processing in more detail, as measurement methodology is important in traffic flow forecasting.

4) The results of the simulations are neither validated by experiments nor supported from experimental evidences. As a consequence, it is difficult to evaluate the reliability of the comparitive analysis here reported. Please add your comments.

5) The Authors omitted the "4. Discussion Authors should discuss the results and how they can be interpreted from the per-spective of previous studies and of the working hypotheses. The findings and their implications should be discussed in the broadest context possible. Future research directions may also be highlighted." Please add your comments.

6) The Authors used a lot of abbreviations, so please kindly add a list of abbreviations to improve the quality of your work.

7) References should be verified and carefully formatting in accordance with the guidelines of the Electronics.

Author Response

Thank you for your corrections, which we have made in the latest submitted manuscript.

(1) We elaborate on the novelty of our paper by comparing it with previous combinatorial models, which we supplement at the end of the Introduction, before the paper organization.

(2) The organization of the paper has been added at the end of the introduction.

(3) We discussed datasets and processing in more detail and added to 4.1.

(4) We enriched our experimental details by adding them to 5.2. The information of a road segment was selected in the two datasets, and the prediction results were visualized, which improved the experimental results and added them to 5.4.

(5) We have added the Discussion section, where we discuss the reasons why our proposed LST-GCN model outperforms other models, and we supplement it at the end of the conclusion regarding future research directions.

(6) List of abbreviations added before references.

(7) References have been checked and revised as required by the electronic guide.

Author Response File: Author Response.pdf

Reviewer 3 Report

The paper proposes a framework named LST-GCN which allows for improving the accuracy of traffic flows prediction 

within the transportation system realm. Precisely, the system combines graph convolutional networks with long-short

term memory networks to take into account Spatio-temporal correlations. 

Initially, the work presents some theoretical background in order to clarify some basic assumptions of the considered 

model. Then, the authors verify the effectiveness of the proposed model through two datasets (PEMS04 and PEMS08) 

which contain traffic collected by several sensors including data such as vehicle speed, occupancy, and traffic. 

The performance of the proposed method has been evaluated in terms of classic indices (MAE, MAPE, RMSE) and contrasted 

with some other techniques including both pure statistical ones (e.g., ARIMA), and ML-based ones (e.g., GRU).

The work seems to be robust and interesting. 

I have just one minor concern pertaining to the time complexity of techniques involved in the proposed model. 

For instance, the “memory” ability introduced by some recurrent structures (such as the LSTM employed for this work) could negatively affect on the time complexity. In this connection, the authors could elaborate more on this point. 

Probably there is no need to perform additional experiments, but the authors could add/comment on some references where such deep approaches have been evaluated both from accuracy and from a time complexity point of view. Some suggestions follow:  

 

- Experimental Review of Neural-Based Approaches for Network Intrusion Management," in IEEE Transactions on Network and Service Management, 2020;

- Network Abnormal Traffic Detection Model Based on Semi-Supervised Deep Reinforcement Learning, in IEEE Transactions on Network and Service Management, 2021;

- Deep Learning for the Classification of Sentinel-2 Image Time Series, IEEE IGARSS, 2019.  

 

Author Response

Thank you for your corrections, which we have made in the latest submitted manuscript. The issue of time complexity is discussed in the Discussion section and references are added.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

I am glad that Authors have responded positively to my comments and corrected the manuscript text in accordance with them. I have no further comments on the manuscript. Good work.