APPLICATIONS

Is your route to work or home a traffic accident hotspot? One researcher at The Ohio State University (OSU) may know. Christopher Holloman, associate director of the Statistical Consulting Service in OSU’s Department of Statistics, has designed a program to identify traffic accident hotspots on Ohio’s roadways. Running on the Ohio Supercomputer Center’s (OSC) Pentium 4 Cluster, Holloman uses SAS software with his program to tell state troopers where fatal and injury accidents, especially those from speeding and drunk driving, are most likely to occur. “We started out evaluating a couple of hundred miles worth of roadway in five major cities in Ohio,” said Holloman. “In Columbus, for instance, we looked at Interstate 270. The Highway Patrol found the information I provided extremely helpful, so it asked me to include all of Ohio.” Holloman said that’s when he turned to OSC. “We already had the code, we just needed a more powerful computer that would fit such a large model,” said Holloman. “Our PCs couldn’t handle the complexity or size of our models for all 88 counties in Ohio. We basically had two options -- either go to OSC or start from scratch and rewrite the program.” Holloman said starting from scratch would make the program more susceptible to errors, not to mention the enormous amount of time it’d take. For instance, Holloman’s program uses a 900-megabyte database that details every traffic accident that occurred on Ohio highways from 2001-2005, and generates 50 gigabytes of output data. The equations that Holloman and his colleagues developed to connect all that data took two weeks to process at OSC. Without OSC’s high performance computing resources, Holloman estimates it would have easily taken him more than a month to process the same data. “Our research involved making predictions for every one mile segment of interstate in Ohio and every five mile segment of U.S and state routes,” Holloman added. “That meant looking at thousands of segments of roadways, which resulted in a massive amount of data.” But it wasn’t just the large number of roadways modeled that made Holloman’s output data set so large. Predictions were made for every roadway under two types of road conditions, good or bad. Also, each roadway had different predictions for each of five different categories of days: Monday through Thursday, Friday, Saturday and Sunday, the day before a long weekend, and holidays. Predictions were also provided by age, which was divided into three groups: under 25, 25 to 64, and 65 and over. In addition, Holloman’s program breaks out results by factors such as alcohol status, speed, and class of vehicle. “Our program relies on reports of injuries and fatalities from the Highway Patrol and incorporates statistics about what makes accidents happen,” Holloman explained. “Common accident causes such as speeding or alcohol consumption are fairly easy to model using OSC’s cluster. Others – such as when a driver will be distracted by a cell phone -- are impossible. So the program makes general forecasts to provide the agency with a report it can use.” Holloman and his colleagues have been issuing reports to the State Highway Patrol in advance of every major holiday since July 4, 2005. That first report only covered interstates around major cities. Last fall, they expanded their computer model to include all interstates, U.S. routes and state routes for which crash data were available. Currently, the Ohio State Highway Patrol is using the program to help position its cruisers during major holidays. The program, however, is publicly available and can be adapted for use by any state. They've also combined the program with Google Earth, which Holloman said makes the tool even easier to use. Google Earth offers an interactive map of the entire globe, including major roadways. The OSU program color-codes the roadways in Ohio, so that users can zoom in to see the general likelihood of accidents in any region of the state. "Everyone would love to be able to predict exactly where and when the next crash would be, but there are just too many factors involved and too much randomness to do that," he said. "We can confidently make broad statements, like whether a particular piece of roadway is riskier at a particular time." For instance, the program indicates that most speeding accidents in Ohio happen during weekday rush hours, and most drunk-driving accidents happen on the weekends between 2:00 and 3:00 a.m. -- after the bars close. In Columbus, most speeding accidents happen on the northern portion of the outer Interstate 270 beltway. But State Route 315, which divides the city north to south, is a hotspot for drunk-driving accidents. "We can make predictions for every major roadway in Ohio, under all possible road conditions, for every hour of the day, for every day of the week," Holloman said. 2006 was the safest year ever on Ohio’s roadways, according to a USA TODAY analysis of preliminary statistics reported to the National Highway Traffic Safety Administration. Approximately 1,238 people died on the state’s roadways last year -- a 6.6 percent drop from 2005. State troopers credit the decline to new research initiatives like Holloman’s. "We see our research as a supplement to officer expertise, and to the efforts of the highway patrol's quantitative analysis group, which does its own analysis of crash data," said Holloman. “It's just one more tool in the patrol's toolbox.” For more information about OSC’s computing resources, visit its Web site.