Why Focus on Road Segments for Traffic Forecasting Models?

[denihilo13]

I'm having a problem with my undergraduate research. My professor wants me to justify why studying a road segment in getting traffic characteristics is better than studying a road network. My objective is to test the effectiveness of a forecasting model in predicting traffic flow. I really appreciate if you could refer research articles that are relevant to my problem. I really haven't found a relevant answer yet. Thank you
P.S. I am not sure if this is the right page for this question :)

 

[jim mcnamara]

How are you defining 'traffic flow'? Do you mean flow at or above saturation? I think maybe you should tell us what your textbook indicates.

 

[denihilo13]

How are you defining 'traffic flow'? Do you mean flow at or above saturation? I think maybe you should tell us what your textbook indicates.

Thank you very much for your reply...I am referring to vehicles at free-flowing condition.So maybe below saturation?

 

[jim mcnamara]

What I am saying is you need specific items to investigate. Flow is vague. How do you define it? Axles per hour? Average vehicle velocity? Queue lengths at an intersection? Intersection wait times? How do you want to measure it? How do you propose to analyze that data?

You have to do this on real world data or on data from some kind of simulation, which it seems you want to do. I cannot help you beyond that. We have zero information for anybody to help you with. You will have to get this mystery value over both a segment and the entire system. And they have to be the exact same type of data or you cannot compare them. You may also have to do some kind of network analysis to make sense of things.

Define traffic flow. Period.

 

[donpacino]

My professor wants me to justify why studying a road segment in getting traffic characteristics is better than studying a road network. My objective is to test the effectiveness of a forecasting model in predicting traffic flow.

What type of project is this (canid homework assignment, masters thesis, phd dissertation, undergraduate project, etc)?
When you say your professor wants you to justify... is that the assignment? Or do you want to study a road segment and he/she thinks its a bad idea?

To test effectiveness of a model you also need a basis of comparison. Get known working models, subject them to the same data, and see what the differences are.

 

[berkeman]

Are any of the items in this Insights article helpful?

https://www.physicsforums.com/insights/analysis-road-capacity/

 

[denihilo13]

What type of project is this (canid homework assignment, masters thesis, phd dissertation, undergraduate project, etc)?
When you say your professor wants you to justify... is that the assignment? Or do you want to study a road segment and he/she thinks its a bad idea?

To test effectiveness of a model you also need a basis of comparison. Get known working models, subject them to the same data, and see what the differences are.

It's an undergraduate project. Yes, I wanted to study the traffic flow (vehicles/time) at a road segment. However, he doesn't agree that studying only segments will do unless I could show him papers that are also studying segments only. I have found several papers doing it i.e. measuring the number of vehicles that flow on a road segment every 15-minutes, however, they didn't give a solid comparison why they studied a road segment over a road network to test the model. Thank you for your reply

 

[denihilo13]

What I am saying is you need specific items to investigate. Flow is vague. How do you define it? Axles per hour? Average vehicle velocity? Queue lengths at an intersection? Intersection wait times? How do you want to measure it? How do you propose to analyze that data?

You have to do this on real world data or on data from some kind of simulation, which it seems you want to do. I cannot help you beyond that. We have zero information for anybody to help you with. You will have to get this mystery value over both a segment and the entire system. And they have to be the exact same type of data or you cannot compare them. You may also have to do some kind of network analysis to make sense of things.

Define traffic flow. Period.

I see. I apologize for not giving much information in my OP. Anyway, as of now I am thinking of comparing several short-term traffic forecasting algorithms (e.g. ARIMA, Kalman Filtering, and smoothing) in forecasting traffic flow (vehicles per 5-mins, vehicles per15-mins, so on...) on a road segment. My objective (like all objectives in the literature) is to test which model has more accuracy when it comes to predicting values multi-step ahead. I don't seem to have a problem with the methodology, model,and reasons why it's an important research. The problem arises when my professor wanted me to justify why I am only planning to study a road segment and not the entire network. He wanted it to be on systems perspective. However, studying a network doesn't seem to fit my objectives which is to mainly test the accuracy of model X in predicting traffic flow (vehicles/time) because I thought(I may be very wrong) that it only requires a road segment(which many others have been doing)to gather data and test the model. Although, he doesn't require me to study the network and the road segment at the same time to compare. He just asked me to find papers that compare both (which is relatively few; I found some but they didn't quite answer my question) and argue it in my paper why its best to study a road segment than a road network(based on what others say) for this specific research. P.S: I just found some research articles yesterday that gave comparisons of road segment and network which quite of answered my questions. They all say that the main reason is that studying the network is very complex (I know it but would it be a valid argument if I told my prof that it is the main reason why it's not much studied in the literature?); taking the road dynamics must be considered etc.

 

[denihilo13]

Are any of the items in this Insights article helpful?

https://www.physicsforums.com/insights/analysis-road-capacity/

Thanks:)

 

[jim mcnamara]

There is something called path analysis. There is an R module for it, for example. It evaluates potentially disparate sets of multiple variables and does a kind of straightforward multiple regression on them. First get sets of data from different road segments, I assume datasets also exist for the network as a whole. Treat the network whole dataset as one large collection of samples then see how or if there is any correlation between individual segments, using PA.
Then run the same PA test between several different single road segments, two at a time. If you are correct in your assertion, this will show it.

Please read this - I am not happy with my description. https://en.wikipedia.org/wiki/Path_analysis_(statistics)

 

[denihilo13]

There is something called path analysis. There is an R module for it, for example. It evaluates potentially disparate sets of multiple variables and does a kind of straightforward multiple regression on them. First get sets of data from different road segments, I assume datasets also exist for the network as a whole. Treat the network whole dataset as one large collection of samples then see how or if there is any correlation between individual segments, using PA.
Then run the same PA test between several different single road segments, two at a time. If you are correct in your assertion, this will show it.

Please read this - I am not happy with my description. https://en.wikipedia.org/wiki/Path_analysis_(statistics)

Thank you very much. I'll definitely try it :)

 

[Ketch22]

A secondary bit of insight.
There are some studies that are available that support your professor. In some cities such as Seattle, some parts of Portland, LA, and portions of downtown Houston ( these are the ones I have personal looked at). The mass flow is significantly different from the segment flow. This is due to Grid lock which is a condition caused by the drivers not by the traffic flow.
If you can show that the challenge is also driver related then you can make a case for segment flow modeling. It is also a case of public education not just accepting the public grumbling.

I guess the overall point from me is that you really need to restrict the research to an issue that can be dealt with in one paper. Your professor will be happier if you address his issue and you will do more, if you can look at a singular point.

 

[denihilo13]

A secondary bit of insight.
There are some studies that are available that support your professor. In some cities such as Seattle, some parts of Portland, LA, and portions of downtown Houston ( these are the ones I have personal looked at). The mass flow is significantly different from the segment flow. This is due to Grid lock which is a condition caused by the drivers not by the traffic flow.
If you can show that the challenge is also driver related then you can make a case for segment flow modeling. It is also a case of public education not just accepting the public grumbling.

I guess the overall point from me is that you really need to restrict the research to an issue that can be dealt with in one paper. Your professor will be happier if you address his issue and you will do more, if you can look at a singular point.

Oh, Thank you very much for your insight :D

 

[CWatters]

I know nothing about this subject so I might be wrong but...

Has it occurred to you that the segment approach might be the best to use and your prof may know it. He might just want you to justify your choice.

 

[256bits]

I just found some research articles yesterday that gave comparisons of road segment and network which quite of answered my questions. They all say that the main reason is that studying the network is very complex

A network would be comprised of segments.
Would not one need to obtain the data from segments first, and then move on to the network analysis.
ie whether or not certain segments are bottlenecks, have more flow than other segments of the network, queues take longer to dissipate at some locations than other, or are more prone to form, etc, etc. That would be near saturation, or above, and you did say free flow. One can have parts of a network free flow and other parts at saturation or above, depending upon inflow, outflow, and other factors that play in at certain points.

Has it occurred to you that the segment approach might be the best to use and your prof may know it. He might just want you to justify your choice.

if you have to give an oral presentation, and then are asked about your choice of segment/network, what is your answer other than "uh... "
Something to think about.