Avatar

Documentation on exception error rates

0

Hi, I'm preparing a manuscript for publication describing our study where we used Geotab GO7 devices to construct a crash risk score (based on driving data and exceptions). I want to be as thorough and transparent as possible when describing our methods and Geotab's data quality. I reviewed the online documentation, and couldn't find information on: 

 

  1. Error exception rates and possible causes for when an exception may be missed. 
  2. In regards to the curve-based algorithm: what is the acceptable 'error-limit' described in the documentation? Also, if to confirm my interpretation is correct, would it be correct to say the algorithm essentially cuts up the data and fits least square lines (within the error-limit) between data segments to form a best fit curve?

Thanks

David Bui

Official comment

Avatar

Hello David,

I apologize for the delay.

1) You are misinterpreting how to curve logic works. The curve logic is used to determine which data we need to send from the device to the server in order to have all the information we need to process exceptions. We also interpolate data between points in order to properly calculate data values between actual logged data points.

The only field validation tests that are done are done by our customers as well as Geotab employees (we all have devices in our own vehicles). We report issues and bugs to our development team and then that gets fed back to the final product as part of a new version or a patch to an existing one. We do not have accurate statistics regarding exceptions we missed but should have caught,

2) I've gotten the firmware team to add the actual error values we use for the algorithm. Please find them here:
https://docs.google.com/document/d/1FDRb841Cp3a7RsovtnTAH6MQxQZzgMs9lv54WsgopAM/edit#

It can be found in the GPS section under Logging.

Please let me know if there is anything else I can do.

Rafael Nazareno

Please sign in to leave a comment.

2 comments

0
Avatar
1) In regards to exception error rates, I was thinking more along the lines about if whether there may be instances where a defined exception should log but does not log. For example, I have a "Speeding 5mph over posted limit for >10s" rule. In theory, if we are logging on a second by second basis, we should never miss any exceptions. However, from my understand, if we use the curve logging algorithm it would effectively reduce the number of logged data points so we would be sampling less than second by second data and could miss logging an exception if the curve algorithm somehow dropped a section of data greater than 10s. Is that correct? Or am I misinterpreting how the curve algorithm works. 
 
Has Geotab done any field validation tests where they committed a defined number of exceptions and reviewed logged exception data to ensure they were logged correctly? For example, go out and commit 5 speeding exceptions and check to see if 5 exceptions were logged. We've had anecdotal reports from some drivers and managers that they've heard the in cab exception beeps,  but did not see an exception logged in the reports.
 
2) The text about the curve algorithm states "In order to give an accurate representation of the data, only the points deemed necessary are kept, which is determined through predefined allowable curve error values."
Could you elaborate on what exactly the 'predefined allowable curve error values' are? This is something I would like to report in our paper. 
David Bui 0 votes