search

nooreendabbish / Traffic

JSM 2016 GSS Data Challenge
1 stars 2 forks source link

How Risk Level for each PAR is Decided?? #21

Closed chenchen715 closed 8 years ago

chenchen715 commented 8 years ago

Hi Patrick:

I just realized one thing, and wanted to understand how you were handling and how we should handle it.

Is our analysis based on each accident, or based on each person? I realized that in my old model, using the GES2013 data you recoded, I was basing the analysis on each person who was involved in the accident. For example, if an accident involves 2 cars, 2 drivers and 2 other passengers, I was counting all that were involved by using GES2013 data.

How are you handling it now? I noticed that you recoded some new data called GES2013.driver. But it still is counting the case (PAR) twice when doing analysis, in previous example where 2 drivers were involved. What do you think?

I was thinking maybe we should make some assumptions, and clean the data before doing analysis. Here are a few assumptions I thought of:

  1. Each case (PAR) involving both drowsy and non-drowsy drivers, it should be considered as a drowsy case.
  2. If a risk parameter for a case (PAR) has more than one evaluations, for example, two cars, one is HEAVY_TRUCK=1 and the other is HEAVY_TRUCK=0, should be counted towards the worst scenario, meaning this case should have HEAVY_TRUCK=1.

Please let me know what you think.

Chen

chenchen715 commented 8 years ago

The model will be built on person (driver) level, and may leave the intra-accident inter-driver correlation as a future research topic.