Near repeat pattern display within HunchLab

For the past few months, the law enforcement team has been developing analysis and risk forecasting features for HunchLab, our geographic crime analysis, early warning and risk forecasting system,  as part of our latest  National Science Foundation research grant.

As we mentioned in a previous blog post, we’ve been working with Jerry Ratcliffe from Temple University to implement a web-based version of his near-repeat pattern analysis.   But what is a near-repeat pattern and what does it mean to you?

Imagine that your neighbor’s apartment is burgled.  Intuitively, you might imagine that, following this event, the burglar won’t strike nearby right away.  But you would be wrong.  Professor Ratcliffe and his colleagues have found that there is a sort of “contagion” effect to some types of crimes, including burglary, shootings and some types of theft.   This means that there is actually a relatively high risk that you will be re-victimized by a burglar in the few days and weeks after the initial event.  In addition, your neighbors are more likely to be victims as well.

It turns out that this “near repeat” risk can be quantified.  So, now, imagine that your local law enforcement agency could describe the elevated risk of subsequent burglaries and thereby better allocate policing resources to help deter further incidents in the area.  The near-repeat pattern analysis lets us determine how far — in both spatial and temporal measures– this contagion effect extends around a crime incident.  It enables us to identify clustering in historic crime incidents across the city and then apply the model we find to current incidents and generate a map of the risk landscape.  These maps can be used to adjust officer deployment strategies to best mitigate risk.

For example, if your residence is burgled within Philadelphia, your risk of subsequent burglaries within the next 14 days goes up by close to 500%.   Similarly, if your neighbors’ house up to 400 feet away from you is burgled, your risk increases by 28% for the next 14 days.

But wait.  There’s more.  A second feature we’ve been working on is called Hunch Focusing.  Within HunchLab, we cover a police department’s jurisdiction with sets of statistical tests (Hunches) that run on a schedule to determine if abnormal patterns in crime incidents have emerged.  When HunchLab detects an abnormality within a specific Hunch it generates an email alert to the appropriate police staff so that they can figure out how to address it.  For people who have seen HunchLab in action, this is not new – these are features that have been available for a few years.  But what if we could automatically refine the parameters of a Hunch to tell the officer even more about the incidents that led to the alert?   Does the pattern extend to other nearby areas?  Is the pattern durable even if users make small changes in how they are testing the data?

We’re working to help law enforcement officials answer these questions by letting the system “focus” the Hunch, sort of like a camera lense that is being adjusted to bring an image into focus.  To do this, we are applying a technique known as “genetic algorithms“.  This approach takes a particular set of parameters and adjusts or “mutates” them slightly and tests the fitness of the results.  We take the initial Hunch and create a collection of mutations in which we change the parameters slightly.  We then take these mutations and breed them among each other to generate a new collection of Hunches from which we select the best ones.  This process of mutation and breeding occurs through several generations until our outcome is a collection of valid Hunches that are much better than our initial Hunch.  They might show that the pattern extends further North from our initial Hunch or that it persists even if we compare against the last 2 years of crime data instead of only the last year.

We’re excited to share what we’ve been working on within HunchLab with you.  We believe that these more advanced tools will help law enforcement agencies improve public safety for us all.  And we are working on more cutting edge improvements that we’ll unveil over the next few months.  Let us know if you would like to see a demo.