For Preventative Automated Driving System (PADS): Traffic Accident Context Analysis Based on Deep Neural Networks

Round 1

Reviewer 1 Report

1. In the abstract, Line 16-17: The statement "when a traffic accident, the offender's injury can be predicted with 85% accuracy and the victim’s case with 67%" - This has the assumption that an accident is always caused by someone and involved two parties. In some cases, there are no offenders: an accident may only involve one party, such as when the accident is caused by hazard conditions and/or poor maneuver.
   The author need to specifically mention the database or the algorithm only has accidents that involves two or more parties and there is always one party to cause the accidents.
2. Line 29-30: for the statement "AVs are defined ... without human assistance": reference is needed for this statement.
3. Line 32: "driverless driving is possible from level 5": Why is it "from Level 5"? There is no other levels above Level 5. Should be "at level 5".
4. Line 44: "that trust is important": Trust of what? Trust of AVs safety, or trust of AVs reliability, or trust on AVs in general? The authors should specify it clearly.
5. Line 50-51: "accidents caused by unpredictable conditions such as black ice, sink-50 hall": in this case, there is no offender or who is the offender?
6. Line 63: "human vehicles(HVs)" should be "human driving vehicles(HVs)"
7. In many places, abbreviations are used without first giving the full terms or definition, such as the VISSIM in Line 66, LOS in Line 68, LSTM, SAE, RBF, SVM, ARIMA in Line 108, DBN in Line 110, R-CNN in Line 121, AP in Line 123, BTF in Line 150.
8. In some places, the full terms or definitions of some abbreviations are unnecessarily used: such as "Level of Service" in Line 145 if the authors had defined it properly when the LOS was first used in previous section. It is the same as "human driving vehicles" in Line 146-147, "artificial intelligence" in Line 148, and "Long-Shore Term Memory" in Line 151, "Preventive Automated Driving System" in Line 160-161, "Deep Neural Network" in Line 163.
9. Line 70: "growing queue": Queue of what? You may mean "queue of traffic". Please make it clear and don't let readers guess what you mean.
10. Line 71: "because it affects HVs": It is not clear about which aspects of the HVs are affected. The authors may need to clarify this.
11. Line 76-79: "investigated ... marginal travel cost": The author only described what the they investigated, but does not give any result. what's the result of this referenced research?
12. Line 94-95: "trolley dilemma problem" - reference is needed.
13. Line 120: "they proposed a new methodology(combining or extension)" - who proposed the new methodology? a new methodology of what?
14. Line 123: "outperforms the state-of-the-art stereo-based method". It is not clear what or which state of the art method is used. Reference is needed. Otherwise, how readers suppose to know what this state of the art method is.
15. Line 126-127: "approximately 85% accurate" - 85% accurate of what? 3D object recognition? or Driving decision?
16. Line 130: "outperformed the state-of-the-art method" - The authors should clearly mention what the state-of-the-art method is.
17. Line 134: Suggest to change "This study means that it is based" to "This study is based".
18. The paper used different reference formats: such as in Line 133 and 127, it used a format of (year) and [index], while more often it used the [index] format. Suggest the author to check all the references over the paper and use only one format.
19. Line 152: "and awareness surveys(acceptance)" - Is it awareness survey or acceptance survey? In the previous sections, there is only acceptance survey is mentioned.
20. About the methodology
    1. suggest the authors to translate the Korean characters to English so most readers can understand the table.
    2. Line 206-207: "the month and day are not relevant to traffic accidents significantly." - Firstly, it depends on how do you define "significant", so at what percentage of accuracy does the author think it is significant? Secondly, accidents in many cases are directly related to or caused by hazard conditions such as adversary weather conditions. And weather conditions are closely related to month or at least seasons, such as snow in the winter and storm in the summer time. How could you say their relevance to accidents is not significant? Or do you have any references to support your statement?
21. About the Design of Optimal Deep Neural Networks(DNNs) in Chapter 4
    1. The authors claimed "optimal deep neural networks". But there is no optimization except tuning the DNN parameters? Is this what the authors mean by optimization?
    2. Line 225-226: for "the Tensorflow and Keras libraries", references are needed.
    3. Line 232, 236, 239, 252, 269: suggest to change "17-18" to "2017-18" as there is no "17-18" in the presented dataset in Table 3.
    4. Line 233: "①set" - what does the "①set" stand for? Where did you give a description about it?
    5. Line 234-235 - "the degree of injury to the offender showed over 80% accuracy": Suggest to change to "the predicted accuracy of the degree of injury to the offender is over 80%." 
    6. Figure 2.a) and 2.b):
       1. suggest to change "data set" to "dataset"
       2. what does the red dash line rectangular stand for?
    7. Line 260: "we simulated based on ①Set" - Simulated what? The sentence does not make sense. Suggest to revise this sentence.
    8. Line 265, 267, 269: "[256,128,64,64]" - The [index] is used for references. You can simply write these numbers as "256, 128, 64 and 64" or other formats. It would be beneficial to describe the meaning of these numbers (although you may mean the number of nodes in each hidden layer).
22. About the results:
    1. Line 275-284: These discussions are based on Table 6 and Figure 4. You should refer to the table and figure at a suitable place at the beginning of this paragraph. Otherwise, readers have to guess what are you talking about?
    2. Line 287-288: "so the vehicle type and the violation of the law are judged to be the main factors determining 287 the degree of injury of the offender" - Somehow, this contradicts with our common sense: does the speed has any effect on accidents and the degree of injury? The authors needs to give a reasonable explanation.
    3. Line 303-304: "while for the victim, the link speed is a more important factor(See Table 7 & Figure 5)." - From both Table 7 and Figure 5, it is obvious that the link speed is a very important factor that affects the degree of offender's injuries (ranked the 2nd highest importance) and the degree of injuries of the victim's injuries (ranked the first important factor). The question here is that if the link speed is so important, why didn't the authors do any model optimization based on the link speed in Chapter 4, and why didn't the author compare the injury accuracy by using link speed as a feature in Table 6? The lack of optimization and accuracy comparison with link speed data as a feature is either something the authors omitted for some reason or a deficit of the methodology in this research. The author should either add more results that related to the link speed, or they should give some explanations about why it is not included in the model optimization and corresponding comparison of the accuracy.
23. English: there are a few grammar errors and many errors in terms of mixed use of past and present tense, plural and single of nouns, lower case and upper case, and others. For example:
    1. Line 15: "a Deep Neural Network technique" should be "the Deep Neural Network technique"
    2. Line 53: "application" should be "applications"
    3. Line 54 and 57: "chapter" should be "Chapter"
    4. Line 64: "Volume by capacity" should be "Volume to Capacity"
    5. Line 92: "accident" should be "accidents"
    6. Line 93: "driver" should be "drivers"
    7. Line 107: "compare ... and ..." should be "compare ... to ..."
    8. Line 115: "area" should be "areas"
    9. Line 228: "set" should be "set at"
    10. Line 244: "Build" should be "Building"
    11. Line 250, 252: "epoch" should be "epochs"
    12. Line 251: suggest to change the "was shown" to "was achieved"
    13. Line 254: "Build" should be "for Building"
    14. Line 269: "hidden layer" should be "hidden layers". Suggest the author to check other places in the paper where it should be "hidden layers".
    15. Line 261: "show" should be "showed"
    16. Line 313: suggest to delete the "need of"
    17. Line 316: "leading" should be "lead", "a difference" should be "the differences"
    18. Line 335: suggest to change "soared to" to a different word. An increase from 81% to 85% can hardly be described as "soared to".

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

1-Speed data is not shown clearly on how they have been collected and the picture is also blurry, which is not clearly presenting how speed data is collected?

2-Wasn't is possible to collect data for more number of years rather than only one year? The better thing was to have at least three year of studies in order to be more confident on analysis? Is it possible if you could consider more number of years in your analysis?

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

1. The reference format: please check the reference format of the MDPI Electronics. If you use [index] format, there is no need to list the the year information. Suggest to remove the (year) info in the references in lines 64, 66, 67, 71, 74, 85, 93, 98, 103, 114, 123, 130, etc. Also please check all other references over the paper.
2. Line 90-91: "2.3. AVs Traffic Accidents and 90 Derived AVs ethics". It seems this part is dangling. The author needs to either remove this or revise the sentence.
3. Fig.2a), 2b): the Y-axis title should be clearly labelled as "Accuracy (%)", or "Injury Accuracy (%)". Please also check Fig.3 and Fig.4
4. Line 273: "a hidden layers" should be "a hidden layer"
5. Line 276, 278, 279, and 281:
   1. The author needs to explain the meaning of the numbers in "256, 128, 64 and 64" when it first appeared in Line 276.
   2. This is just a discussion and suggestion: somehow, the representation of ""256, 128. 64 and 64" hidden layers" is still awkward in the text. If these numbers represent the number of nodes in each hidden layer, I would write it as "...accuracy was 84.15% with (256,128,64, 64) nodes in the hidden layers.". You may use other ways you think it is most suitable.
   3. If you adopt the suggestion in point b) as above, please also consider to change them in Table 6.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf