Page 227 - Big Data Analytics for Connected Vehicles and Smart Cities

208 Big Data Analytics for Connected Vehicles and Smart Cities Practical Applications and Concepts for Transportation Data Analytics 209

neck that is exhibiting multiple speed drops and increases around the reference
speed represent a single bottleneck.
Unfortunately, the speed variability, bottleneck, and characterization
work discussed above did not yield a statistically significant difference between
the conditions before the VSL sign implementation and afterward. This led to
discussion and consideration of alternative approaches to the analytics.
Speed variability and bottleneck characterization techniques were applied
as part of the original analysis. Unfortunately, this did not yield a statistically
significant result that showed an improvement from the before situation to the
after situation. Speed variability on its own did not provide a useful measure of
the effects of the VSL signs.
At the midway point of the work, some six months into the schedule, a
briefing and discussion with the client provided a breakthrough insight into the
evaluation problem. In a discussion regarding the subjective effects of the VSL
signs, a senior member of the state DOT staff stated:

When you drive the corridor during incidents conditions, you get the
impression that the traffic is somehow more tranquil than before we
installed the variable speed limit signs.

This insight led to a review of the parameters that were adopted for the
evaluation. It became clear that simple speed variability, standard deviation,
or averages may not provide insight addressing the subjective statement. The
evaluation team then considered the word tranquil and realized that the op-
posite of tranquil would be turbulent. This thought led to the determination
that an evaluation parameter that measured traffic turbulence may be a more
appropriate measurement and might yield a clear result.
Subsequent discussions with the data science team identified a new can-
didate evaluation parameter—traffic turbulence. Traffic turbulence was defined
as the change in speed between adjacent segments times that occurrence of that
event. Further analysis also identified that the most significant location to mea-
sure traffic turbulence is at the end of the queue, where it would be expected
that the VSL effects would be most pronounced due to the warning given to
drivers approaching the end of the queue.
This represented a discovery moment. While the previous analytics work
did not yield a statistically significant result, it did form the basis for under-
standing the characteristics of the data sufficient to form the foundation for a
traffic turbulence analysis. Given a new understanding of the complexity of the
analysis, the team decided to focus on traffic turbulence analysis based on traffic
speeds, while putting the other factors to one side.