Keiran explored and trained various logistic regression models to estimate the likelihood that an individual would be victimised. His first model asserted that all people in its sample set would be victimised. It was correct 70% of the time. regression model types (including decision trees, random forests and neural networks) looking for improvements on that benchmark.
After including information about previous brazil rcs data victimisation in the past 12 months, Keiran’s model at the end of the weekend was able to make predictions with 89% accuracy.
Simon White, from Cardiff University, has been developing his work on identifying expressions of fear of crime through Twitter for comparison with CSEW data. He was able to provide us with a presentation and a cut of his data.
At the hackathon I presented the work I’d done on a project called A Cop on Every Corner, conceived by Alex Blatchford (Digital Policing, Metropolitan Police). The idea is to pair spare moments police officers have (between major taskings, or as they travel) with micro-taskings – small jobs they can voluntarily accept to increase policing presence, and clear up smaller issues that might otherwise fall between the cracks.
As you can imagine, with fleetingly small amounts of time you have to have a good idea of what tasks will have the best impact, and how to easily identify them.