Project description

Traffic disruptions often begin with a crash or a visual distraction, and then the disruption quickly spreads to other parts of the road network. Slowdowns at some locations are especially influential, causing associated slowdowns that affect a large swath of regular vehicles and road freight. These are locations where problems are not necessarily more frequent, but where problems tend to spread widely to other roads. Existing traffic forecasting approaches, including deep neural nets, find correlations between traffic at different locations, but do not find cause-and-effect relationships. Likewise, current work on crash impact analysis is inadequate, because slowdowns come from more than just crashes, and because a slowdown can have subtle, far-reaching effects, such as through rubbernecking. The objective of this project is to find the road locations that are most influential at causing slowdowns at other road locations. Finding these means we can target road improvement funds at the most problematic locations, which will make the road network more robust and help road freight move more efficiently. Our expected results will consist of a method to find the most influential roads, with an analysis of California traffic to show which parts of the road network are most prone to causing widespread slowdowns.