Stay connected

Get the latest insights into Canadian business


Finding patterns in spatial data

By Dr. Kim Rossmo | January 9, 2014
Finding patterns in spatial data
Dr. Kim Rossmo

Connecting geospatial analyses of criminal investigations to data analytics in business is not the stretch it may first appear to be. Both police and corporate officials must make decisions under conditions of uncertainty. Data can reduce uncertainty, but first it must be analyzed and developed into information and knowledge.

Once you look behind the Hollywood image, a criminal investigation simply consists of turning data into knowledge. A police detective collects and analyzes evidence in order to determine what happened and who was responsible. An investigation has two basic functions: find the offender; and prove his or her guilt. In some cases, the first stage may generate thousands of potential suspects. Detectives, therefore, need a reliable system for prioritizing searches under conditions of information overload.

Crime sites are carefully examined for DNA, fingerprints, tire marks and other physical evidence. However, the fact that the crime site’s location is also a clue is sometimes forgotten. Moreover, in a serial murder, sexual assault or arson case, there are multiple sites that, when considered together, produce a larger geometric pattern. The examination of this pattern is called geographic profiling – the analysis of a series of crime sites to determine the most likely area of the offender’s residence. The purpose of a geoprofile is to provide an optimal search strategy and an information management tool for suspect prioritization.

Geographic profiling was originally developed from my PhD research at Simon Fraser University’s School of Criminology in British Columbia. I was a member of the Vancouver Police Department at the time, so I tried to combine criminological theory with my background in mathematics and police experience. Understanding the operational requirements of detectives (users) and the nature and type of information available in a criminal investigation (task) was central to this effort. To help meet the need of police agencies for a suspect prioritization tool, Environmental Criminology Research Inc. (ECRI) was formed to develop an operational geographic profiling software system. In 1997, ECRI released the first version of Rigel, named after the star in the constellation Orion the hunter. The RCMP was the first customer.

The integration of the software with the analytic methodology of geographic profiling was important to the development of Rigel. We tried to accomplish this through the adoption of a few key design principles.

Using a common foundation of readily available data

  • Input is limited to data typically available in a police investigation.
  • A scientific method for crime linkage was developed to better identify a crime series.
  • A robust methodology is used to minimize sensitivity to missing and erroneous information.

Following an analytic process

  • The division of analytic labour recognizes that some tasks are best done by computers and others by humans.
  • The process is theory-based and, therefore, adaptable to new circumstances; assumptions are clearly articulated to prevent misapplication.
  • A high-quality training program helps users maximize the system’s potential.

Focusing resources and getting results

  • Databases frequently contain address information; geography is, therefore, a more useful prioritization metric than other offender traits.
  • The capability of the model to be applied in a wide variety of situations is just as important as its precision.
  • The ability to visualize and manipulate the geoprofile helps users understand the results.

It turns out that several disciplines have a need to find lost or yet-undiscovered things. Over recent years, this has led to a number of innovative applications of geographic profiling. Zoologists and biologists have used the technique to study great white shark foraging off the coast of South Africa, bat-hunting patterns in Scotland, bumblebee search patterns, and the spread of invasive algae in the Mediterranean. The military is interested in applications for counterinsurgency, and the intelligence community for counterterrorism. Epidemiological researchers have used the model to locate mosquitogenic water sources from malaria cases in Cairo and contaminated water pumps from cholera deaths in 19th-century London. Currently, we are exploring its use for hunting cyber-attackers and maritime piracy bases. Other suggested but still untested uses include archeological excavations, mineral exploration and business marketing.

Data turned into information and knowledge can be very powerful, although this process must be followed by logical and unbiased thinking. Developing methods of finding the best paths through data jungles and cognitive mazes are important and necessary undertakings. Identifying patterns that others have missed in a mass of data provides a key advantage both in public safety and in business.

Information overload?

The case of Operation Lynx in England illustrates the problem of information overload in a serial crime inquiry

In 1996, English detectives investigating a series of abductions and rapes had recovered DNA and a partial fingerprint. However, the offender was not in their national database, and the print was too fragmented for a computerized search. Investigators began a manual review of fingerprint files, but this effort was so labour intensive that a prioritization scheme was required. English police asked the Vancouver Police Department was asked to prepare a geoprofile, using the rape sites and locations where the offender purchased items with a credit card that had been in a stolen car connected to one of the attacks. The results of the geoprofile centred on two neighbourhoods (see Figures 1 and 2). After a few months of searching, a match was found, and police arrested the offender. He lived in the peak of the geoprofile, in one of the identified neighbourhoods; his mother lived in the other. During the manhunt, detectives initiated 24,324 actions, tested 2,177 DNA samples and entered 33,628 names into their computer system. Given the large regional scope of the investigation, the manual fingerprint search would not have been successful without a narrowed geographic focus. 


The content of this field is kept private and will not be shown publicly.
About PwC Canada Hide Footer