Waze data for urban interventions - Part I

5 minute read

This is the Part I of a series of three posts. If you’re interested, check out Part II or skip directly to Part III.

The challenge

In 2017, the Municipality of Joinville established a partnership with Waze through their open data Connected Citizens Program, in which they offer an API for municipalities to access real-time data from their local urban traffic.

Never before have cities had access to real time traffic data to support decision making in the urban planning process and it was a privilege for me to be leading this initiative in my home town. We named the project Smart Mobility.

One of the main challenges of such an open ended opportunity is to determine what kind of data analyses can bring the most value to one’s organization. It’s easy to get overwhelmed by the plethora of possible paths and end up with no really useful analysis. To avoid this pitfall, we’ve gathered the traffic engineers of our division to map out the process of going about implementing urban interventions in the city. The output can be seen below:

Figure 1: Smart Mobility urban intervention process.

And then brainstormed in which steps would the data at hand be valuable:

Figure 2: Steps of the intervention process that can be enhanced with data.

The data

Waze provides a data stream that is updated every two minutes. It’s basically a JSON file containing relevant information about every single traffic jam in the city which, in a nutshell, contains the following features:

  • datetime
  • traffic jam unique identifier
  • traffic jam length in meters
  • traffic jam speed in km/h
  • traffic jam delay in seconds (delay caused to the driver when compared to free flow)
  • coordinates vector (a list of cordinates indicating the geographical position of the jam line)

Now, following the methodology described in Figure 1, let me present how we used the data to create value for our urban mobility management.

1 - Diagnostic

Here our goal is to identify the most critical streets and prioritize them using an objective criteria. This is useful not only to better use our limited budget and human resources, but also to convince external stakeholders (legislators, civil organizations, the midia, etc..) that we’re addressing the right problems.

But what makes a street critical? There are three independent metrics that could be used for this assessment:

  1. How slow are the traffic jams when they happen in that street;
  2. How long do the traffic jams last when they happen in that street;
  3. How often does that street experience congestion throughout the day.

To simplify our analysis, we grouped the three above criteria as following:

  1. Traffic Criticality
    • How slow are the traffic jams when they happen in that street
  2. Traffic Probability:
    • How long do the traffic jams last when they happen in that street
    • How often does that street experience congestion throughout the day

Using Waze’s data to calculate these metrics, we ended up with the figure below:

Figure 3: Traffic Probability vs Traffic Criticality.

These are very interesting results! On the top left region of the graph we have streets that experience very heavy traffic once in a while - those are the streets to which many people require interventions based on anecdotal claims such as “last week I was stuck for 30 minutes in that street! This city’s traffic system is completely saturated!”.

On the bottom right, we see streets that are almost always congested, but without causing much delay for the drivers and are probably not worth the investment for an intervention.

Finally, on the top right corner of the graph, we see the critical zone - streets with high probability and high criticality of congestion, that really demand interventions and are worth the investment.

Laying down the streets this way objectively rank them according to a reasonable criteria that avoids investments based on anecdotal or purely political reasons.

2 - Conception of Solution

Here, the data does not really provide us with useful insights. The process of coming up with a solution for a traffic problem still relies heavily on the experience and creativity of our engineers.

3 & 4 - Demand Modeling and Simulation

Great traffic simulation tools are currently available to help assess the effectiveness of a traffic intervention. All of them, however, rely on a good estimate of the traffic flow (cars per hour), sometimes called demand. There are several mathematical models that can help convert point estimates from manual counting or induction loops to a city-wide directional estimate of traffic flow, but given the dynamic nature of the phenomenon, those estimates are quite often very poor.

Waze’s data provide city-wide, real-time data on traffic, but the information does not include traffic flow. We could, nonetheless, use Waze’s data to improve our traffic flow estimate in three ways:

  1. Cross-reference the Waze’s data with our 100 traffic radars spread over the city that do count traffic flow. We would then have a good amount of labeled data that could be used as input to a regression model. An initial analysis suggested that this approach wouldn’t yield good results;
  2. Use Waze’s data to compare against the results of the simulation model. If both differ substantially, more (or better) traffic counts should be carried out to improve the simulation. This approach helps, but does not improve the process significantly;
  3. Use Waze’s data as input to an optimization model that estimates traffic flow in such a way that minimizes the difference between the model’s and Waze’s traffic indicators. This research is being currently being conducted by the University of Sao Paulo, based on the data and analyses that are presented here. More on this topic in future posts.

5 - Results Monitoring

As mentioned before, urban traffic interventions are often the target of severe public scrutiny and criticism. Just as important as the existence of objective criteria to justify its implementation is the presence of metrics to validate its success. To this end goal, Waze’s data is spot on and ready to be used.

Figure 4 shows the average delay due to traffic in five streets that were affected by a controversial intervention implemented in Joinville in July/2018, before and after the intervention. This simple chart was used as main argument to bring objectivity to the table and settle endless discussions with multiple stakeholders that disagreed with the intervention, including local media and legislators from opposing parties.

Figure 4: Average delay before and after a traffic intervention, in five streets.

The improvement was undeniable and the speed with which the government mustered public approval was arguably unprecedented for such a major traffic change.


Leave a comment