Using Machine Learning to Predict Annual Average Daily Traffic Counts

PiinPoint Traffic Counts Methodology

Annual Average Daily Traffic (AADT) counts are used to understand the average number of vehicles that travel on a road segment each day. AADT is often counted by municipalities in order to drive local decision making, and is heavily relied upon by businesses to get insights into traffic patterns of an area. However, because of inconsistent collection methodologies between  municipalities,these studies are not always up-to-date, or available.
‍
Using mobile location data and machine learning, PiinPoint is able to provide accurate predictions of the number of vehicles travelling along a given road segment, as well as the direction of travel, for any road in North America. With a median accuracy of over 81%, PiinPoint’s data instills confidence and offers insights you can trust to make business decisions.
‍

Why PiinPoint has created our own Traffic Count Predictions

Consistent coverage across North America

It’s impossible to make decisions about your business that you can be confident with when you’re relying on outdated dataset or missing key information. AADT are typically commissioned by thousands of different municipalities, each of which may have a different methodology making it impossible to accurately compare the data.
‍
Combining mobile data and machine learning allows PiinPoint to provide consistent coverage across North America, giving peace of mind when comparing traffic counts from region to region and enabling an apples-to-apples comparison regardless of what market you're in. 
‍

Always Up-to-Date

With PiinPoint’s traffic count predictions, you can trust that our AADT counts are always up to date. We deliver insights based on historical traffic patterns across the previous year, rather than looking at a single snapshot in time, allowing you to account for seasonality and keep a pulse on the latest traffic realities.

Predictions you can Trust

We have gone through rigorous accuracy testing and improvements to be confident that we are providing our clients with accurate AADT estimates across North America. Our accuracy measurements come in at a median accuracy of 81.62% for our AADT predictions. 

Always learning and improving

PiinPoint’s Mobile Traffic Counts are always improving. Our machine learning model is constantly analyzing additional data, as well as being supervised by our team of data scientists and GIS developers to further improve its accuracy. 

AVG Vehicle Volume on Segment

40,602

vehicles per day

21,921

Left-Side of road

19,311

Right-Side of road

How PiinPoint Traffic Counts Work

PiinPoint uses machine learning and mobile data to create a proprietary model that estimates the AADT volumes for any given road segment. Machine learning allows PiinPoint to identify trends and patterns within the data, in order to continually train, test and improve our traffic count predictions. 

Training Data

Mobile Traffic Data

PiinPoint has partnered with Near, a leading provider of anonymized mobile location data, to draw insights from 45 million mobile devices each month across North America. 
‍
The mobile location data is completely anonymous and represents a sample size of 3-12% of the population. Near gets the data from over 1,000 iOS and Android apps by asking the user to enable access to their location settings or make use of cookies. 
‍
Using machine learning, PiinPoint is able to extrapolate on the sample of anonymous mobile traffic data to represent 100% of the North American population. PiinPoint takes into account demographics, population statistics, and observation data to ensure accuracy.

Annual Average Daily Traffic (AADT) Data

Testing data for PiinPoint’s prediction accuracy comes from a sample of over 4100 AADT counts along road segments across North America. Covering a variety of road types, locations, geography's and conditions. 

Creating the PiinPoint Traffic Count Prediction Model

The AADT counts are used as a source of truth, and compared against the available mobile data points along the same road segments during the same time period. These road segments are then analyzed to understand variation in the road segment, such as road type, proximity to urban core, and other factors that influence the number of estimated vehicles collected. This allows PiinPoint to understand the variations in the road network and where potential sources of error may occur, as well as do validation testing to measure the accuracy of our estimated vehicle counts compared to the validation AADT data. 

• Direction of traffic
• Number of lanes
• Type of road
• Density of the urban area
• Number of occupants in the vehicle
• Demographics

Data you can count on

PiinPoint calculates accuracy in 2 ways. First, by assessing how closely the predictions match actual AADT counts, and secondly by doing a regression analysis to measure how well the predictions fit the actual AADT counts.
‍
The graph below helps to illustrate how PiinPoint measures our traffic predictions against AADT data. In this scatter plot, PiinPoint was able to achieve an R(2) value of 0.8698, showing that these predictions are highly accurate. The r(2) value is a measurement of error, and the closer to 1 it is, the less error there is in our predictions to the true values. Visually, it is apparent that there are very few outliers within our predictions as well, giving PiinPoint confidence that AADT calculations are highly accurate. 

Figure 1. PiinPoint was able to achieve an R(2) value of 0.8698 when calculating the accuracy of traffic count predictions.