Visualizing predictive crime maps

Posted on 02/10/2011 by


Crime mapping has been a tool employed by police for more than a century, but only recently has the data collection been organized and consistent enough to accurately aggregate spatial crime statistics. I was taking a look at a 1997 paper on predictive crime mapping using artificial neural networks—a topic I recently became very interested in—to predict and visualize crime data.

Olligschlaeger (1997) incorporated an artificial neutral network into a GIS-based early warning system for street level drug markets in the Pittsburgh Bureau of Police. A typical model consists of a number of processing units sending signals to one another through weighted connections (Kroese and Van der Smagt, 1993), where the authors institutes a learning algorithm to shape the network architecture.

What is interesting about using neural networks for spatial statistics is the particular emphasis on complexity in prediction. Using an analogy to cellular automata as a chaotic system, Olligschlaeger notes that:

[t]hey differ from other chaotic systems in that they act on discrete space or grids rather than a continuous medium such as a surface. In a cellular automata machine, each frame (representing all cells in a population) is replaced by a new one according to a specific “recipe,” or rule, in the next epoch (Toffoli and Margolus, 1987). A key determinant of cellular automata rules is how each cell is influenced by neighboring cells.

As such, she uses a ‘game of life’ learning algorithm to analogize grid-based spatial complexity.From this algorithm Olligschlaeger derives a set of choropleth maps (a type of map where areas are shaded in relation to some statistical variable, here drug calls for service per month).

This offers another good example of both data visualization being used outside of an academic context, as well as a greater methodological point about the power of pattern recognition.  Our brains are better at reading patterns when it comes to a complex system (such as time series spatial crime data) than interpreting regression analysis.  I know there is more out there in this field, but this is a particularly good example of data visualization taking advantage of pattern recognition. 

Andreas. 1997. M. Artificial Neural Networks and Crime Mapping. In Crime Mapping and Crime Prevention, Crime Prevention Studies Volume 8. Eds. David Weisburd and Tom McEwen. New York: Criminal Justice Press. 314-345.

Cohen, J., W.L. Gorr, and A.M. Olligschlaeger (1993). Modeling Street-Level Illicit Drug Markets, Working Paper #93-64. Pittsburgh, PA: H.John Heinz III School of Public Policy and Management, Carnegie Mellon University.

 Toffoli, T. and N. Margolus (1987). Cellular Automata Machines: A New Environment for Modeling, MIT Press Series in Scientific Computation. Cambridge, MA: MIT Press.

Kroese, B. J. and P.P. Van der Smagt, (1993). An Introduction to Neural Networks. Lecture Notes. Amsterdam, NETH: University of Amsterdam.