CRIMINOLOGISTS increasingly rely on “geographic profiling”, an examination of the sites of dastardly deeds that narrows down the possib...
CRIMINOLOGISTS increasingly rely on “geographic profiling”, an examination of the sites of dastardly deeds that narrows down the possible identities of serial criminals. A mathematical refinement of this idea has implications for pre-empting both terrorist activity and the progression of infectious disease. To demonstrate it, and to help settle a mystery, Steven Le Comber of Queen Mary, a college of the University of London, has used it to suggest the identity of Banksy, a prolific but reclusive street artist, an example of whose work may or may not be pictured above.
Dr Le Comber, a biologist, learned of geographic profiling from Kim Rossmo, a criminologist at Texas State University, and reckoned it could be useful for epidemiology, too. Although many existing models can predict the spread of disease, Dr Le Comber wanted to look at the past—to identify, say, a breeding site responsible for a malaria outbreak. This geographical backtracking is just what Dr Rossmo had been doing with criminals. Now the pair collaborate on methods which will help pinpoint both crooks and critters.
Their system, Dirichlet process mixture modelling, is more sophisticated than the criminal geographic targeting (CGT) currently favoured by crime-fighters. CGT is based on a simple assumption: that crimes happen near to where those responsible reside. Plot out an incident map and the points should surround the criminal like a doughnut (malefactors tend not to offend on their own doorsteps, but nor do they stray too far). The Dirichlet model allows for more than one “source”—a place relevant to a suspect such as home, work or a frequent pit stop on a commute—but makes no assumptions about their number; it automatically parses the mess of crime sites into clusters of activity.
Then, for each site, it calculates the probability that the given array of activity, and the way it is clustered, would result from any given source. Through a monumental summing of probabilities across each and every possible combination of sources, the model spits out the most likely ones, with considerable precision—down to 50 metres or so in some cases.
The duo have tested the technique before, examining data about the postcards of Otto and Elise Hampel, two anti-Nazi activists who distributed them anonymously in Berlin in the early 1940s. It picked up the couple’s home and Mr Hampel’s workplace, as well as two underground stations where they changed trains.
To try to pin down Banksy’s identity, they gathered the locations of 140 of his alleged works in London and Bristol. Within the peaks they found, as they report this week in the Journal of Spatial Science, were a pub, playing fields and a residential address in Bristol, and three addresses in London. All were associated with one person: Robin Gunningham. This was not, perhaps, much of a surprise. Mr Gunningham was publicly fingered as Banksy in 2008. But this independent result suggests those fingers were pointing in the right direction.via the economist
No comments