If you place a candy bar in front of a young boy, you can predict that he’s going to eat it. And if it’s cloudy outside, you might be able to predict that it’s going to rain. But is there any way that you can predict where a car crash is going to happen?
Turns out, you can — by using predictive analytics. And that’s exactly the technology that Tennessee is tapping to promote traffic safety and prevent major accidents from happening.
Under the “Crash Reduction Analyzing Statistical History” (C.R.A.S.H.) program, Tennessee Highway Patrol (THP) officers factor in data points ranging from festivals and sporting events to weather patterns and areas with a history of accidents to create predictions of where they should deploy their resources.
The program spans five-by-six-mile squares, providing predictions for four-hour periods each day.
“So it might show that between 6 and 10 p.m. the probability of a serious crash is 68 percent in this block,” said THP Colonel Tracy Trott.. “And that’s where the captain should direct his resources.”
The goal of the six month-old program is to give officers a heads up to help prevent crashes or to be nearby should one happen. While nothing is 100 percent accurate, C.R.A.S.H. has been right 72 percent of the time since its inception.
“You have some days when the predictions are right on, and other days when they’re way off,” said Beth Rowan, a THP statistical analyst working on the program. “Mainly what you want to look for is whether the performance of the model is acceptable. And collectively, it’s been very good.”
Traffic fatalities have dropped around 5.5 percent from this time last year, leading officials to view C.R.A.S.H. as being effective, according to Trott.
The software has the capability to factor in any data point while also dismissing points that may not be pertinent. THP has also implemented a model to focus on drivers under the influence of alcohol or drugs, using informatoin like the locations of vendors who sell alcohol.
“The model itself goes through and identifies, if you will, what the most important characteristics are,” Rowan said. “You put everything in that you can, and the model tells you what is important and what’s not.”
The entire program cost THP $243,000, with funding provided through federal grants from the Governors Highway Safety Office. While Tennessee is not the only state beginning to implement predictive analytic software, only a handful of states are using it for traffic patterns, according to IBM’s public safety specialist Mike Reade.
“Oftentimes veteran law enforcement officers will be making those predictions themselves when they’re in the field,” Reade said. “What we do is put a lot of data and fact behind it. The volume of factual data we’re using can’t be done by a human. You need an analytical tool like this to sift through the volumes of data — years of traffic data — to come up with this type of foresight.”