Experiments associated with DataKind's Vision Zero initiative with Seattle Department of Transportation and the University of Washington.
There are three key focus areas: Exposure Estimation To better understand the safety, or danger of specific locations in the city for pedestrians and cyclists, we must understand the relationship between crashes and the total volumes of automobiles, pedestrians and bicycles present at specific locations (i.e. exposure). While automobile volumes are the most commonly collected data of this type, coverage is still not complete. Furthermore, bicycle and pedestrian counts are much less common and cover a much smaller number of city streets. Your mission? Develop separate estimates of automobile, pedestrian and cyclists volumes with the count data available for each group, coupled with information on the built environment, demographics, road network characteristics and other data for these streets. Modeling Crash Probability Understanding what streets and locations are the most dangerous, and what the is probability of a crash at suuch locations can provide policy makers and engineers with actionable information for developing and implementing interventions. Pedestrian and cyclist crashes at specific locations can be characterized as rare events. Therefore, developing models that can provide the probability of crash, even at locations where a crash has yet to occur, is extremely valuable. This analysis can also provide insights into the locations that are the safest and highlight potential characteristics of these locations that contribute to overall safety. Your mission? Using the collision data set, available exposure data (traffic, bike, pedestrian counts), road network characteristics, built environment characteristics, and other datasets to develop models. Traditional approaches to this problem include Negative Binomial, Poisson, Bayesian and Linear Regression models - participants are encouraged to apply these, and/or other statistical/ML approaches for addressing this problem. Modeling Influencing Factors for Specific Crash Types There are different categories of crashes for both pedestrians and cyclists based on how the collisions occurred. Examples of such types include: Right turn, left turn, head on, and rear end collisions. Understanding the contributing factors and predictive variables associated with these specific crash types can elucidate specific interventions that could be implemented to reduce the number of such collisions. Furthermore, similar to the objective above, this work can highlight specific locations where specific crash types are more prevalent. Your mission? Using the collision data set, available exposure data (traffic, bike, pedestrian counts), road network characteristics, built environment characteristics, and other datasets to develop models.