What is the system trying to predict?
The IBM Crime Insight and Prevention software is built on other IBM technology — like IBM Cognos Business Analytics and SPSS Advanced Analytics. It appears that some version of the technology predict crimes, whereas other versions (see discussion regarding Miami and Memphis below) work to predict perpetrators’ identities. IBM’s system can also determine the likelihood of recidivism.
What data are the system's predictions based on?
IBM’s software can use historical crime data, profiles, maps and typology, as well as other events like weather, holidays, or paydays, as input data. It’s unclear which data is required and which is voluntary, though it’s likely that the historical crime data is required.
Overall, IBM claims its systems can bring together “a wide range of data, including unstructured data, for analysis, modeling and prediction,” suggesting no specific limit on what data will or won’t be used. In fact, in describing how their predictive analytics works, IBM says that “all available relevant information must be collected. And we do mean all.”
Thoug there are no specific limits mentioned, there are some specific types of input data IBM uses:
- Historical crime incidents: location, crime type, severity, victims, suspects, convictions, criminal behaviours and attributes. This can come from sources like computer-aided dispatch systems (CAD) or records management systems (RMS).
- Enabling factors: weather, temperature, time of year, month or week
- Trigger events: holidays, festivals or paydays
- Unstructured data: including, images, audio, video and text contained in incident reports, witness statements, suspect interviews, tip information, calls for service, e-mail, blogs, and chat room activity. According to IBM “[t]his information is critical for analyzing interactions and uncovering the attitudes, desires and motivations of criminals to get at the reasons behind crimes. Understanding the ‘why’ helps prediction go beyond assuming that past actions and behaviors will dictate future ones.” It’s likely that this last element plays into a separate IBM system that predicts potential perpetrators/suspects for a crime.
What approach does the system se to make its predictions? How is the data analyzed?
IBM’s crime prediction and prevention system relies on IBM SPSS (originally standing for Statistical Package for the Social Sciences). It’s unclear what exact technique the crime and prevention system uses, as IBM SPSS “offers a variety of modeling methods taken from machine learning, artificial intelligence, and statistics.”
According to IBM, their “[p]redictive modeling does not make the assumption that what has happened in the past will continue into the future. Instead, the model combines analysis of past events with a higher level of logic to determine the arc of events—such as criminal behaviour or crowd patterns or the evolution of organized crime structures.”
How is information presented to users?
Our research has so far been unable to definitively find what this tool looks like in the field. We do know that in some instances (like in Memphis) IBM’s tools can be paired with ESRI ArcGIS — an advanced mapping software.
It’s likely it looks something like this, but we cannot be positive:
How accurate are the predictions? How is accuracy defined and measured?
We’re not aware of any independent analyses of this system’s efficacy. We’re also unaware of how IBM internally defines predictive accuracy for its system.
Despite this, IBM has made specific claims about a decrease in crime after their systems were deployed — like a 27 percent decrease in serious crime overall in Memphis, and a 28 percent decrease in the robbery rate over a year in Manchester, NH. Also, though it’s not a specific claim as to accuracy or crime reduction, a 2012 IBM commercial did seem to provide an implicit claim. The commercial highlights IBM’s predictive analytics, showing a police officer deployed through a predictive system just before a would-be robber arrives.
Which departments are using the system?
IBM gives several examples of where their tools are being used:
- Richmond PD
- Edmonton PD
- Durham PD
- Memphis PD (CRUSH Initiative)
- Miami-Dade PD (see below for more)
- Manchester, NH PD
According to IBM
Blue PALMS, or Predictive Analytics Lead Modeling Software, leverages crime analytics to break cold cases, and catch repeat offenders of crimes. When a crime is committed, an officer collects information that is captured in the Blue PALMS model. Blue PALMS uses advanced analytics to generate a list of potential suspects based on match probability. This list is then delivered to investigators to narrow their focus from thousands of known offenders to those with the highest probability of having committed the crime. Blue PALMS provides stronger leads to investigators by leveraging historical crime patterns and offender modus operandi from huge volumes of data.
How are departments using the system?
See above on Miami-Dade PD. More to come.