Blog-2-Website-Featured-Image.png

AI-based Clustering to Better Understand Corridor Behavior

October 11, 2024

If you’re unfamiliar with AI-based clustering, it’s probably more integrated into your daily life than you even realize. Take your smartphone for example. If you’re an iPhone user, you will have likely interacted with the People and Pets feature inside the photos app. This feature automatically creates albums for different individuals in your image library. But how does Apple automatically differentiate your friend from your dog from your grandfather? It uses an unsupervised clustering algorithm that groups together similar upper body and facial characteristics (a process Apple explains further in one of its machine learning research articles.) AI clustering has countless applications; in the world of traffic management, these use-cases range from accident detection to congestion pattern detection and forecasting.

People and Pets feature inside the Photos app of apple device

Figure 1: People and Pets feature inside the Photos app of apple device

At Axilion, whenever we take on a new project, we first like to gain a deeper understanding of the corridor’s traffic patterns. We do this by performing a traffic cluster analysis referencing the most recent three months of detection data available (corridors with live-data integration capabilities undergo a different process outlined later in this blog). Our clustering algorithm performs an unsupervised AI process that automatically groups similar traffic volumes together, as well as directions of travel, throughout the length of the day. This process generates a series of charts, like the one shown below, that help us visualize and analyze time-of-day traffic patterns across individual intersections and corridors as a whole.

Traffic cluster chart_Tallahassee corridor

Figure 2: Traffic clustering chart for 3.5-mi corridor in Tallahassee, FL (August-October, 2023)

The image above presents Axilion’s cluster analysis for a 3.5-mile traffic corridor in Tallahassee, FL with ten signalized intersections. The X-axis represents the time of day, while the Y-axis indicates the date. Each color corresponds to a different traffic volume, ranging from light (yellow) to heavy (dark purple). This chart focuses exclusively on Monday through Friday, as non-working days have been excluded from the analysis.

What does this cluster diagram tell us?

This diagram provides a clear view of typical traffic patterns along the corridor. We can observe that traffic generally follows a consistent daily pattern: light traffic overnight, followed by a morning rush hour around 7:30 am, which tapers off by 9 am. Although traffic remains relatively heavier throughout the day, it peaks between 3:15 pm and 6:30 pm, with the Friday afternoon peak occurring slightly earlier. Additionally, the analysis highlights some anomalies, such as the period marked by the red box, which corresponds to a hurricane and Labor Day weekend. During this time, traffic patterns deviated from the norm, likely because fewer people were commuting to work.

What can traffic management practitioners do with these insights?

Traffic engineers can use cluster analyses to make signal timing adjustments and better plan for seasonal signal timing decisions. They typically set different traffic signal timing plans for different times of day, each with a unique cycle length tailored to the rise and fall of traffic volumes. However, these time-of-day plans are not always up-to-date or well-aligned with the reality of the traffic behavior. Traffic clustering can help engineers better-align their time-of-day signal changes with the real traffic patterns occurring. Additionally they can better understand seasonal variations in traffic patterns, and schedule different plans for different seasons.

How Axilion uses cluster analysis

In addition to using clustering for understanding corridor traffic patterns, Axilion integrates the clustering capabilities into its X Way platform. When a user generates a new signal timing plan in X Way, they have the option to select the time of day (TOD) plan that best aligns with a corridor’s traffic patterns. In the platform screenshot below, the user can decide between selecting (1) “In-field TOD plan” – maintaining the time-of-day plan currently set up by the agency, or (2) “X-Way recommended TOD plan” – utilizing the time of day plan recommended by X Way based on the traffic patterns revealed in the cluster analysis.

Creating a new signal timing policy_optimization candidate

Figure 3: X Way platform screenshot: Selecting TOD plan when running an optimization

Agencies that have integrated their Advanced Traffic Management System (ATMS) with the X Way platform benefit from near-real-time traffic clustering. Every 15 minutes, X Way can process fresh detection data to compare against historic traffic patterns and guide the recommended course of action for signal retiming.

Concluding thoughts

In conclusion, AI clustering offers a powerful tool for traffic practitioners to gain deeper insights into traffic patterns throughout the day. By leveraging these insights, engineers can make more informed decisions for advanced planning, signal adjustments, and even near-real-time traffic management. The ability to align signal timing plans with actual traffic behavior, as well as adapt to seasonal variations, makes AI clustering an invaluable asset in optimizing traffic flow. At Axilion, we’re excited about the potential of our X Way platform to incorporate real-time data through integration with Advanced Traffic Management Systems (ATMS), further enhancing our ability to respond dynamically to traffic conditions.

About the Author:

Yuval Regev

Yuval Regev

AI Algorithms Engineer

Yuval Regev holds a bachelor’s degree in computer science from Bar Ilan University and has worked at Axilion for two years. Specializing in AI algorithms, Yuval is responsible for data acquisition, analysis, and generating actionable insights to optimize traffic signal timing. Yuval’s work includes performing traffic cluster analyses, creating the input for Axilion’s digital twin simulation, and validating the accuracy of the digital twin.