Chapter 1: Precrime Data Mining


Chapter 1: Precrime Data Mining

1.1 Behavioral Profiling

With every call you make on your cell phone and every swipe of your debit and credit cards, a digital signature of when, what, and where you call or buy is incrementally built every second of every day in the servers of your credit card provider and wireless carrier. Monitoring the digital signatures of your consumer DNA-like code are models created with data mining technologies, looking for deviations from the norm, which, once spotted, instantly issue silent alerts to monitor your card or phone for potential theft. This is nothing new; it has been taking place for years. What is different is that since 9/11, this use of data mining will take an even more active role in the areas of criminal detection, security, and behavioral profiling.

Behavioral profiling is not racial profiling, which is not only illegal, but a crude and ineffective process. Racial profiling simply does not work; race is just too broad a category to be useful; it is one-dimensional. What is important, however, is suspicious behavior and the related digital information found in diverse databases, which data mining can be used to analyze and quantify. Behavioral profiling is the capability to recognize patterns of criminal activity, to predict when and where crimes are likely to take place, and to identify their perpetrators. Precrime is not science fiction; it is the objective of data mining techniques based on artificial intelligence (AI) technologies.

The same data mining technologies that have been used by marketers to provide personalization, which is the exact placement of the right offer to the right person at the right time, can be used for providing the right inquiry to the right perpetrators at the right time, before they commit crimes. Investigative data mining is the visualization, organization, sorting, clustering, segmenting, and predicting of criminal behavior, using such data attributes as age, previous arrests, modus operandi, type of building, household income, time of day, geo code, countries visited, housing type, auto make, length of residency, type of license, utility usage, IP address, type of bank account, number of children, place of birth, average usage of ATM card, number of credit cards, etc.; the data points can run into the hundreds. Precrime is the interactive process of predicting criminal behavior by mining this vast array of data, using several AI technologies:

  • Link analysis for creating graphical networks to view criminal associations and interactions

  • Intelligent agents for retrieving, monitoring, organizing, and acting on case-related information

  • Text mining for examining gigabytes of documents in search of concepts and key words

  • Neural networks for recognizing the patterns of criminal behavior and anticipating criminal activity

  • Machine-learning algorithms for extracting rules and graphical maps of criminal behavior and perpetrator profiles



1.2 Rivers of Scraps

"It's not going to be a cruise missile or a bomber that will be the determining factor," Defense Secretary Donald Rumsfeld said over and over in the days following September 11. "It's going to be a scrap of information." Make that multiple scraps, millions of them, flowing in a digital river of information at the speed of light from servers networked across the planet. Rumsfeld is right: the landscape of battle has changed forever and so have the weapons—if commercial airliners can become missiles. So also has how we use one of the most ethereal technologies of all human creativity and imagination: AI.

AI in the form of text-mining robots scanning and translating terabyte databases able to detect deception, 3-D link analysis networks correlating human associations and interpersonal interactions, biometric identification devices monitoring for suspected chemicals, powerful pattern recognition neural networks looking for the signature of fraud, silent intrusion detection systems monitoring keystrokes, autonomous intelligent agent software retrieving e-mails able to sense emotions, real-time machine-learning profiling systems sitting in chat rooms: all of these are bred from (and fostering) a new type of alien intelligence. These are the weapons and tools for criminal investigations of today and tomorrow, whether we like it or not.

Which of the 1.5 million people who cross U.S. borders each day is the courier for a smuggling operation? Which respected merchant on ebay.com is about to abandon successful auction bidders, skipping out with hundreds of thousands of dollars? What tiny shred of the world's $1.5 trillion in daily foreign exchange transactions is the payment from an al-Qaeda cell for a loose Russian nuke? How many failed passwords attempts to log into a network are a sign of an organized intrusion attack? Finding the needles in these types of moving haystacks and the answers to these kinds of questions is where data mining can be used to anticipate crimes and terrorist attacks.