Real-Time Propensity Factor

Close

Ferrovial Business Lines Projects Real-Time Propensity Factor United States 2019 Project Year $1.5MM Innovation Investment Payback between 2-4 years Our highways in Texas feature dynamically priced Managed Lanes toll system or Express toll lanes, which make it possible for drivers see how rates are adjusted dynamically based on average speed or the number of vehicles at any given time. The Real Time Propensity Factor project aims to leverage Machine Learning techniques and the tools available to understand real drivers preferences beyond average speed or average congestion propose adjustments to the tolls on the Managed Lanes accordingly. The RTPF is a new layer on top of the existing Toll Setting Algorithm, and it considers the variations or anomalies in customer’s propensity to use the Express Lanes under similar travel time savings, toll rate, and time of day but with different external/environmental conditions. The machine learning algorthms are able to detect the anomalies and new external explanatory factors for those anomalies that could not have been identified otherwise. Thanks to the Real Time Propensity Factor, an adjustment to the price is suggested based on all available environmental variables anytime the market share on the MLs is expected to be abnormal due to those external factors. Finding More Efficient Solutions Cintra is committed to finding more efficient solutions thanks to Big Data and innovation, increasing its resources and investment in data analytics year after year. That’s why, back in 2019, the Data Analytics team was able to identify anomalies in traffic capture rates, analyze those anomalies and create a new algorithm to optimize the  solutions and improvements points. It works this way: Currently, thanks to the existing systems, a fee is calculated each five minutes depending on the density and speed of circulation, to guarantee drivers always a minimum speed of 50mph at the Express Lanes . The new algorithm works on top of the existing Tolling Algorithm identify anomalies as other variables come into play for each particular location of the highway (% of trucks, congestion at the decision point etc). Each location has its own model that has been adjusted after detecting changes on the propensity to use the express lanes from variations in those other factors. We have just begun installing the workflow in our production environment so we can run the entire process (anomaly detection, clustering, and choice of toll factor) on both historical and real-time data. We are exploring approaches to incorporating the new algorithm to the standard Tolling Algorithm in place, ultimately absorbing many of the its real-time functions and moving us from a parametric to a flexible, data-driven approach. Value of the Solution It’s well known the traffic density at Dallas-Fort Worth urban area. Since the construction of the “LBJ” highway 40 years ago, it has become one of the most saturated roads in the United States. A similar story occurred on the NTE corridor in Ft. Worth. The managed lanes toll system was designed to ensure there are always express lanes available for a fast and reliable trip. However, it is important to price this new product in accordance with the value perceived by the customer and that does not always coincide with the contractually mandated measurements of average speed and congestion for each segment. RTPF gives the operators the option to adjust the prices in accordance with the real preference of the customer based on an extended set of variables , particular to each location, beyond the traditional ones.

Medium