Dr. Colleen McCue, president and CEO of MC2 Solutions, gave a talk at VCU entitled “What can Amazon and Wal-Mart teach us about fighting crime in a recession?” She talked about using data mining and analytics tools to not just fight crime, but prevent it.
I really enjoyed McCue’s talk, since she had some really, really cool examples of how she used analytical tools (data mining, statistics, and a little optimization) to fight and prevent crime. (I am also a CSI jumkie).
For example, McCue used data mining to determine which armed robberies escalated to an aggravated assault. One challenge was that these events are rare (3% escalate) and data is unreliable, since it is mostly provided by eyewitnesses (If you’ve seen CSI or Law & Order once or twice, this should be no surprise). Most of the useful data is in a narrative, which is hard to process in an automated way. Once this challenge was met, the data were used to proactively located police officers in the riskiest areas (areas that were most likely for a robbery to escalate to an aggravated assault), rather than locate police areas with the highest rate of (non-escalating) robberies. It turned out that one of the riskiest areas was where suburbanites drive into the city to buy drugs (as one police officer put it), where they were then used as “walking ATMs.” Some resisted being robbed, and were subsequently shot.
McCue used a similar method to develop a risk-based method for locating and scheduling police officers on New Year’s Eve. Her method reduced random gunfire by 47%, increased gun seizures by 246%, and saved $15K. The key was noting that most of the violence in New Year’s Eve occurs in a two hour window around midnight–and being able to convince decision-makers to radically change shift assignments.
In her talk, McCue noted that most of the time, data analysis confirms what is already known. But sometimes, it identifies new things. For example, Wal-Mart was able to identify that before storms hit people stock up on bottled water and strawberry Pop Tarts. Her approach is similar, with McCue’s Pop Tarts being non-intuitive ways to fight crime. After scouring non-acquaintance rape data, for example, McCue was able to determine that prior property crime offenses (not prior assault or peeping Tom offenses) was the most likely indicator for rape.
The most interesting thing I learned about the talk is about how behavioral analysis should be used in Numb3rs-esque ways to fight crime. For example, segmentation uses behavioral analysis, much like marketing, since it, in effect, targets certain populations. It makes no sense to look at homicide, McCue argues. Instead, she looks at drug-related homicide, manslaughter, etc. McCue developed a segmentation tool to determine motive for homicides, in order to help develop a list of suspects early on. Once a list of likely suspects are formed, the perpetrator can be found. Her model only takes data that would be available early in an investigation, since it is critical to make progress on a case (with 48 hours) before it goes cold. Finding likely suspects is one of the most critical parts of an investigation. Apparently, behavior models are more accurate than models that rely on the “hard data,” since hard data is unreliable whereas behavior is homogeneous. Being a numbers nerd, this makes me uncomfortable (I could barely type that last sentence), but I impressed with the results.