GeoPython 2021

Predicting Traffic Accident Hotspots with Spatial Data Science
2021-04-22, 14:00–14:30, Track 1

Road traffic accidents are a major health and economic problem worldwide. Spatial Data combined with Data Science tools and models can help anticipate high-risk locations dynamically based on factors such as traffic, weather, and road signaling.


Road traffic injuries are among the ten leading causes of death worldwide and they have a significant effect on the world’s economy. Governments and the private sector are making big efforts to reduce these numbers and, as a result, today we can have real-time information on traffic and weather conditions, in addition to traffic statistics. However, this information is available either post-accident or it is static. Knowing where accidents happen and the conditions under which they happen is very powerful information that can be leveraged to identify hotspots dynamically and take action to anticipate accidents (e.g., city administrations can share this information with their citizens and organize their traffic police accordingly, and logistics companies can use this information to avoid specific routes)

In this talk, we will show how different spatial data sources (road traffic, weather, road signaling, human mobility, points of interest, and working population) affect traffic accidents and how they can be used to identify hotspots dynamically. First, I will walk you through the spatial support selection phase and the process of bringing all data sources to the same support. Once the data is ready to be consumed, I will walk you through a spatial and temporal analysis of accident data using different tools and techniques. This first analysis will already give us some hints on typical characteristics of traffic accident hotspots. I will then present a predictive model using Regression Kriging with Random Forest as regressor that will allow us to predict annual accidents. This predictive model will help us validate our hypothesis of changing conditions affecting traffic accidents and the potential of defining dynamic hotspots. The analysis focuses on the city of Barcelona (Spain), which has a rich Open Data catalog available.