Predicting future crime activity utilizing machine learning and Flickr metadata

James White


Supervised by Steven Schockaert; Moderated by Irena Spasic

Flickr photos and social media in general, are increasingly used to understand and analyse what is happening in the world around us. For this purpose, it is especially useful that many Flickr photos have explicit meta-data associated with them, such as geographical coordinates, a time stamp, or (in the case of Flickr) descriptive tags.

The aim of this project is to relate publicly available crime statistics to Flickr metadata associated with photos taken in a particular demographic. This information can then be interpreted to suggest and predict future crime rates/ crime shifts across the nation in specific spatial regions. For example, a region with a known, high pick-pocketing crime rate, may have recent Flickr descriptive tags suggesting the area is busy, tight, & hectic, and therefore this crime rate is destined to stay relatively consistent.

Initial Plan (02/02/2014) [Zip Archive]

Final Report (06/05/2014) [Zip Archive]

Publication Form