Onboard navigation and mobile applications can tell drivers how to avoid traffic jams. Trouble is, most of the drivers are already on the road, perhaps already in the jam. But IBM is about to deploy a system that will predict traffic flow up to an hour before it occurs, giving travelers ample time to avoid trouble.
During pilot tests in Singapore, forecasts made across 500 urban locations accurately predicted traffic volume 85 to 93 percent of the time and vehicle speed 87 to 95 percent of the time. Similar results were achieved in Finland and on the New Jersey Turnpike.
The key to success is predictive modeling—software that combines real-time data from road sensors and cameras, as well as GPS transponders in taxis, with historical traffic information, roadwork conditions and weather forecasts. Each week the model recalibrates based on statistics from the most recent six weeks. It broadcasts advisories to electronic road signs and car navigation displays. The system also predicts when a congested road will return to normal flow.
On supporting science journalism
If you're enjoying this article, consider supporting our award-winning journalism by subscribing. By purchasing a subscription you are helping to ensure the future of impactful stories about the discoveries and ideas shaping our world today.
IBM has signed contracts with two U.S. transportation authorities to deploy a full system, according to spokesperson Jenny Hunter. The locations will be announced soon. Singapore may commit as well and is also testing a variation that will predict bus arrival times for riders waiting at bus stops.
In each location, ongoing work will optimize the advisories. If, for example, Highway 1 is clogged and too many drivers who receive messages flock to Highway 2, it will become clogged; engineers will customize the model so it can determine whether sending the messages to only 25 or 40 percent of drivers, say, would best balance the two roads. And because a high percentage of drivers now carry cell phones, IBM is working with several telecom companies to be able to track the continually changing density of their phones along roadways, which could provide finer-grained modeling. To protect privacy, the identity of individual phones would not be disclosed.
The company has also announced its intention to develop services that could tell individual subscribers ahead of time which of various routes would get them to their preselected destination fastest, given current conditions. Voice recommendations would be sent to a person’s vehicle navigator or cell phone.
Similar analytics are being applied to other applications. “The beauty of predictive modeling is that it translates across disciplines,” says Robert Morris, IBM’s vice president of service research in Armonk, N.Y. For example, the company’s laboratory in Haifa, Israel, is testing a program called EuResist that predicts the success of different drug cocktails for an HIV patient over time. The software analyzes the person’s HIV genotype and his or her current health characteristics against an evolving database of treatment outcomes for more than 33,000 patients and 98,000 therapies. Similar applications might determine which type of breast or prostate cancer treatment could benefit a patient most.
IBM is also working with the Washington, D.C., Water and Sewer Authority to predict in real time where problems are likely to arise—such as which sewer lines might flood during storms. The goal is to adjust valves ahead of time, systemwide, to minimize overflows and to deploy maintenance crews to specific locations early. In March, IBM opened a predictive analytics lab in Xi’an, China, to help business clients such as Xi’an City Commercial Bank anticipate customer trends before they occur.