Prediction of Freeway Traffic Breakdown Using Artificial Neural Networks

Round 1

Reviewer 1 Report

This paper presents an algorithm for prediction of traffic breakdown (i.e., the transition from an uncongested to a congested state of traffic flow) on freeways based on artificial neural network (ANN).

The structure of the paper is good, but the related work should be moved from Section 1 and 2 to a new Section 2 (roughly from page 2 line 52 to page 2 line 78 and entire Section 2.2). It could be also extented. Other papers predicting traffic behavior (not only traffic breakdown) and other papers predicting other similar things using similar ANNs are good candidates.

The entire approach is well described and seems to be sound. However, it would be better to use multiple sites as the basis for the preparation of the algorithm (for example the feature selection). The usage of a single site potentially limits the usabilit of the algorithm for other sites. Alternatively, the developed algorithm should be tested at different sites after it was developed based on the single site. Yet another possibility is to discuss the changes, which would be necessary to adapt the algorithm for a different site.

The future work is missing at the end of the paper.

The references seem to be relevant and relatively up to date. However, there are only two papers from 2020s, which could indidate that the author missed the latest research in this area.

The figures are appropriate and of sufficient quality.

The English is very good, the number of errors and typos is quite low.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

The paper entitled: "Prediction of freeway traffic breakdown using artificial neutral networks" presents interesting point of view on traffic beakdown prediction. The ANN machine learning method used to predict the occurrence/or not of a traffic breakdown is correctly described and correctly applied. I discuss the sense of its application, in substance, to this issue. In transportation network design, the causes and identified situations of breakdowns are of interest in order to be able to prevent them or minimize the duration of the resulting traffic congestion. The ANN method used makes it possible to determine, with a high degree of accuracy (Tab.6.), whether the situation will occur under the given input conditions covering the state at moment t - which is also the moment of forecasting. So:

1.  what are the possibilities of practical application of the model?

2. Who can use it and how?

This should be clearly articulated in Section 4 "Results and discussion".

Other concerns and comments:

3. Tha data covers 6074 samples of which breakdowns were observed in only 38 cases. Overlaying this with the division of the collection into training and testing, there were few cases of breakdowns in these collections. What impact does this have on the effectiveness of the ANN model?

4. Why is the division into training and test collections not random but 2 months - one month? This may affect the effectiveness of modeling. Each period can be distinctive in some way. In order to prevent the influence of the specificity of the time period, a random method is used to select subsets. 

5. Due to the many limitations in the applicability of the model on other data, I propose to include a section: "Limitation" before section 5.

6. In the introduction section it is worth adding information about the increase in emissions when driving in traffic jams, which is an ecological consequence of the lack of traffic flow. You can read about the functional (deterministic) relationship between speed and traffic density at doi:10.7409/rabdim.017.011. It is worth responding to these conclusions in the context of the considerations made in the manuscript.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

Thank to the Authors for carefully reviewing my comments and explaining them in detail. I accept the arguments given on the two main issues (comments 1 and 2). The other additions are also satisfactory. The completed manuscript is clearer and more attractive to the reader.