Predicting Perceived Cycling Safety Levels Using Open and Crowdsourced Data

Authors: Jiahui Wu*, University of Maryland - College Park, Lingzi Hong, University of Maryland - College Park, Vanessa Frias-Martinez, University of Maryland - College Park
Topics: Urban Geography, Geographic Information Science and Systems, Transportation Geography
Keywords: cycling safety, spatio-temporal analysis, open data
Session Type: Paper
Day: 4/3/2019
Start / End Time: 2:35 PM / 4:15 PM
Room: Harding, Marriott, Mezzanine Level
Presentation File: No File Uploaded


Cycling communities have been related to lower obesity rates and lower stress levels. Nevertheless, one of the main obstacles to increase ridership in cities is the lack of information regarding perceived cycling safety at the street level. City planners have typically used extensive road network and traffic information to approximate cycling safety levels. However, this approach requires the deployment of expensive sensors thus making it hard for many cities to get access to accurate cycling safety maps. In this paper, we evaluate several methods to predict urban cycling safety at the street level, exclusively using public information from open and crowdsourced datasets. We also present an open-source, crowdsourced platform developed to help cities gather ground truth cycling safety labels so as to train their own local models to achieve the highest safety prediction accuracies. We evaluate the proposed approach in the city of Washington D.C. and achieve F1 scores of 66%, 70% and 88% when five, four or three different cycling safety levels are considered.

Abstract Information

This abstract is already part of a session. View the session here.

To access contact information login