Autonomous Short-Term Traffic Flow Prediction Using Pelican Optimization with Hybrid Deep Belief Network in Smart Cities

Round 1

Reviewer 1 Report

This paper presents a method based on pelican optimization and hybrid deep belief network to predict traffic flow. Overall, the paper is well-written, and the results are nicely presented. My main concerns are as follows;

- Abstract can be organized better. The first few sentences can be removed or shortened.  Some key results should be included in the abstract.

- Related works section should be improved. Only a  few previous studies have been included. I think there are numerous studies on short traffic flow prediction in general.

- Why do the authors claim that the proposed algorithm is better than the existing ones?

- Section 3 - Were real data used in this study? A brief introduction to the used traffic data should be provided.

- Section 3.1 - Why data normalization is important or required here. How the max and min values were set? Was any time duration considered for this normalization? Furthermore, this normalization process could highly depend on the selected time window (for future prediction).

- Fig. 4 - What is meant by "time index" in this figure? What was the prediction time window? In conclusions (line 369) “near future prediction” is mentioned. What would be the optimal time window for this prediction?

- It is not clear what ‘traffic flow’ was predicted here. Was it the network traffic flow or the traffic flow on a road section? Probably a separate figure can be provided OR Fig. 1 can be improved to describe such.

Author Response

This paper presents a method based on pelican optimization and hybrid deep belief network to predict traffic flow. Overall, the paper is well-written, and the results are nicely presented. My main concerns are as follows;

- Abstract can be organized better. The first few sentences can be removed or shortened.  Some key results should be included in the abstract.

As per the reviewer comment, the abstract is precisely rewritten in the revised manuscript. Kindly refer Page 1, Abstract.

- Related works section should be improved. Only a  few previous studies have been included. I think there are numerous studies on short traffic flow prediction in general.

Based on the reviewer comment, the recent references are included in the revised manuscript. Kindly refer Page 3, Lines 123-139.

- Why do the authors claim that the proposed algorithm is better than the existing ones?

Thank you for the comment. We claimed that the proposed model has obtained better results over the other existing models. A detailed comparison study shown that the proposed model has gained effective prediction results compared to other models.

- Section 3 - Were real data used in this study? A brief introduction to the used traffic data should be provided.

As per the reviewer comment, the experimental details are provided in the revised manuscript. Kindly refer Page 9, Section 4, Lines 294-302.

- Section 3.1 - Why data normalization is important or required here. How the max and min values were set? Was any time duration considered for this normalization? Furthermore, this normalization process could highly depend on the selected time window (for future prediction).

Thank you for the comment. Data normalization is a set of techniques and rules used to improve the consistency of data, standardize it, and maintain its integrity. The process removes any redundancies or duplicates in the database and reduces the storage space requirements. In this work, the min-max data normalization approach is employed, which normalizes the data into a range of [0, 1].

- Fig. 4 - What is meant by "time index" in this figure? What was the prediction time window? In conclusions (line 369) “near future prediction” is mentioned. What would be the optimal time window for this prediction?

Thank you for the comment. We have referred ‘time lag’ as ‘time index’ in the study.

- It is not clear what ‘traffic flow’ was predicted here. Was it the network traffic flow or the traffic flow on a road section? Probably a separate figure can be provided OR Fig. 1 can be improved to describe such.

As per the reviewer comment, the experimental details are provided in the revised manuscript. Kindly refer Page 9, Section 4, Lines 294-302.

Reviewer 2 Report

The article entitled Autonomous Short-Term Traffic Flow Prediction using Pelican Optimization with Hybrid Deep Belief Network in Smart Cities is a very interesting article with a current topic. The neural network and the optimization method are clearly described.

Please go through the article carefully and explain each abbreviation.

The authors state that Fig. 1 depicts the overall process of the proposed method. If this is already the case, then it is necessary to explain it in more detail. Especially in the part of data collection, which goes into the optimization process and which is the temporal and spatial reach of the traffic load prediction. 

In the explanation of The Proposed Model, a sentence is written: Traffic flow prediction is using present traffic flow data, historic, and other related is stated statistical data for establishing an appropriate mathematical method.  I ask the authors to explain in detail in the context of the proposed model how the mentioned data were used.

Figure 1 shows the intersection, and table 2 shows the measured results for the section (one section). It is not really a clear concept what is the subject of prediction in the context of temporal and spatial scopes. In which unit is the traffic load expressed? This information will give an insight into the time range of the prediction, and it is not specified anywhere.

The authors state in the title of the paper smart cities. The paper does not refer to that term anymore, nor what it means in the context of the model. Considering that it is in the title of the article, the term smart cities would have to be explained in detail in the context of the article's content.

Explain the RBM technique in more detail. Explain the abbreviation.

Figure 2 shows the structure of the neural network. Please explain the input parameters of the input layer. If the traffic load is the subject of prediction, why is there more than one neuron in the output layer?

Table 1 shows data for MAPE analysis. Explain in one sentence what MAPE analysis is.

What is more important, please describe the database against which the prediction results are evaluated (best, average, and worst). What is measured data? Where are they measured? Intersection, section? How long? How big is the base database? Is the prediction data compared with field data in real-time, and how? Functional road level?

If the described proposed model is applied in real traffic conditions, the description of the experiment is completely missing, and a detailed description of the application of the model in a specific experiment is essential for the scientific contribution of the article.

If the proposed model is applied to the future modeled conditions of a smart city, then specify in detail, which network segment model frameworks were used and which simulation/microsimulation model was used. How the traffic model of the future demand of the future smart city is set up and how the experimental settings can be replicated?

A detailed description of the practical application of the proposed model should also explain what the actual data in table 2 mean.

Table 3 and Figures 5-8 compare the results of the proposed model with 5 other models that are not shown or described anywhere in the article. They were not even mentioned in the existing research chapter. It is not clear whether the specified models were applied to the same experimental network segment and which traffic flow indicator was selected for comparison.

Explain what is meant by Lags (Table 4).

If all of the above is explained, then the scientific contribution will be clearer. A detailed description of the experiment is an important segment of the article in understanding and interpreting the results.

 

The conclusion is written too generally, and the scientific contribution is not argued and critically highlighted. In the concluding chapter, describe the difference between the proposed model and the existing methodology, which must be explained beforehand. Explain how the recommended model can improve the traffic conditions of the particular observed network segment or network model. Describe carefully the advantages, disadvantages, and limitations of the recommended model and in which segment it can be improved in future research. Describe in which experimental environment it gave good results, and in which it should be further tested.

Author Response

The article entitled Autonomous Short-Term Traffic Flow Prediction using Pelican Optimization with Hybrid Deep Belief Network in Smart Cities is a very interesting article with a current topic. The neural network and the optimization method are clearly described.

Please go through the article carefully and explain each abbreviation.

As per the reviewer comment, we have improved the language quality of the manuscript and thoroughly proofread for grammatical as well as typographical errors.

The authors state that Fig. 1 depicts the overall process of the proposed method. If this is already the case, then it is necessary to explain it in more detail. Especially in the part of data collection, which goes into the optimization process and which is the temporal and spatial reach of the traffic load prediction.

Thank you for the comment. We have properly discussed the overall working of the proposed model in the revised manuscript. Kindly refer Page 3, Last paragraph.

In the explanation of The Proposed Model, a sentence is written: Traffic flow prediction is using present traffic flow data, historic, and other related is stated statistical data for establishing an appropriate mathematical method.  I ask the authors to explain in detail in the context of the proposed model how the mentioned data were used.

As per the reviewer comment, the above-mentioned issue is corrected in the revised manuscript. Kindly refer Page 3, Last paragraph.

Figure 1 shows the intersection, and table 2 shows the measured results for the section (one section). It is not really a clear concept what is the subject of prediction in the context of temporal and spatial scopes. In which unit is the traffic load expressed? This information will give an insight into the time range of the prediction, and it is not specified anywhere.

The authors state in the title of the paper smart cities. The paper does not refer to that term anymore, nor what it means in the context of the model. Considering that it is in the title of the article, the term smart cities would have to be explained in detail in the context of the article's content.

As per the reviewer comment, necessary information is given in the revised manuscript.

Explain the RBM technique in more detail. Explain the abbreviation.

As per the reviewer comment, RBM technique is clearly given in the revised manuscript. Kindly refer Page 5, Lines 171-176.

Figure 2 shows the structure of the neural network. Please explain the input parameters of the input layer. If the traffic load is the subject of prediction, why is there more than one neuron in the output layer?

As per the reviewer comment, necessary information related to the neural network is given in the revised manuscript. Kindly refer Page 5, Lines 181-185.

Table 1 shows data for MAPE analysis. Explain in one sentence what MAPE analysis is.

As per the reviewer comment, the definition of MAPE is given in the revised manuscript. Kindly refer Page 9, Lines 303-306.

What is more important, please describe the database against which the prediction results are evaluated (best, average, and worst). What is measured data? Where are they measured? Intersection, section? How long? How big is the base database? Is the prediction data compared with field data in real-time, and how? Functional road level?

If the described proposed model is applied in real traffic conditions, the description of the experiment is completely missing, and a detailed description of the application of the model in a specific experiment is essential for the scientific contribution of the article.

If the proposed model is applied to the future modeled conditions of a smart city, then specify in detail, which network segment model frameworks were used and which simulation/microsimulation model was used. How the traffic model of the future demand of the future smart city is set up and how the experimental settings can be replicated?

As per the reviewer comment, the experimental details are provided in the revised manuscript. Kindly refer Page 9, Section 4, Lines 294-302.

A detailed description of the practical application of the proposed model should also explain what the actual data in table 2 mean.

Table 3 and Figures 5-8 compare the results of the proposed model with 5 other models that are not shown or described anywhere in the article. They were not even mentioned in the existing research chapter. It is not clear whether the specified models were applied to the same experimental network segment and which traffic flow indicator was selected for comparison.

Explain what is meant by Lags (Table 4).

If all of the above is explained, then the scientific contribution will be clearer. A detailed description of the experiment is an important segment of the article in understanding and interpreting the results.

The conclusion is written too generally, and the scientific contribution is not argued and critically highlighted. In the concluding chapter, describe the difference between the proposed model and the existing methodology, which must be explained beforehand. Explain how the recommended model can improve the traffic conditions of the particular observed network segment or network model. Describe carefully the advantages, disadvantages, and limitations of the recommended model and in which segment it can be improved in future research. Describe in which experimental environment it gave good results, and in which it should be further tested.

Based on the reviewer comment, the key findings and possible future works are included in the conclusion section.  Kindly refer Page 15, Section 5.

 

Round 2

Reviewer 1 Report

The authors have addressed the comments provided by the reviewer. I have no further comments. 

Reviewer 2 Report

I have no additional remarks.