Data Mining for Juvenile Offenders

Suppose that a fifteen-year old girl is convicted of theft.  After time in the juvenile justice system, the girl is obliged to participate in a rehabilitation program.  If she lives in Florida, a data mining program will suggest a particular program for her.  Florida State Department of Juvenile Justice recently announced that will be using IBM predictive analytics software to reduce recidivism.  According to IBM, the software helps identify at-risk youth who “can then be placed in programs specific to the best course of treatment to ensure offenders do not re-enter the juvenile justice system.”  It analyzes “key predictors such as past offense history, home life environment, gang affiliation, and peer associations to better understand and predict which youths have a higher likelihood to re-offend.”  The Vice President of predictive analytics at IBM explained that “[p]redictive analytics gives government organizations worldwide a highly-sophisticated and intelligent source to create safer communities by identifying, predicting, responding to, and preventing criminal activities.  It gives the criminal justice system the ability to draw upon the wealth of data available to detect patters, make reliable projections, and then take the appropriate action in real time to combat crime.”  Florida’s hope is that if youth are “rehabilitated with effective prevention, intervention, and treatment services early in life, juveniles will not enter the adult corrections system.”

Providing well-tailored rehabilitation services to juvenile offenders like our fifteen year old can be valuable if it indeed delivers on its promise.  Nonetheless, as states further automate important government decision-making in this way, they ought to address potential troubling uses of the program’s recommendations.  Mission creep is surely possible.  Although initially used to recommend a rehabilitation program, the program’s findings could be shared with other agencies.  For instance, a finding that a youngster has a high likelihood of recidivism may impact that person’s educational opportunities.  The United Kingdom’s Ministry of Justice uses IBM’s analytics software to “assess the likelihood of prisoners re-offending upon their release to help improve public safety.”  It is unclear if the Ministry employs the data mining program to tailor rehabilitation programs or if law enforcement instead has more far-reaching use, such as the monitoring of recently released offenders apart from their parole.  Michael Pinard has done important work on the damaging collateral consequences of convictions: flagging someone as likely to re-offend has great potential to add to that problem.

You may also like...

1 Response

  1. Dr. leff says:

    This is not new. People have been talking about using techniques such
    as regression to predict recidivism and help judges make good sentencing
    Many states used very specific guidelines–I had a programming
    course with many students from my University Law Enforcement and Justice
    Administration Departments. I had the students program the guidelines
    as a real-world example of a complicated if statement.)
    rediction, sentence the juvenile to one randomly and watch the outcomes.
    d be ank going back to the sixties show that statistical approaches
    predict recidivism better than “clinical experts.”
    In six studies, statistical approaches less sophisticated than modern
    data mining, did as well as experts. In 1973, the National
    Advisory Commission on Criminal Justice Standards and Goals
    recommended the use of statistical methods in sentencing decions.
    (Caroll, John S., “Judgments of Recidivism risk: Conflicts between
    Clinical Strategies and Base-Rate Information”,
    Law and Human Behavior
    volume One, Number Two 1977,
    page 191-198.

    One question is whether the more sophisticated techniques of machine
    learning and data mining are doing a better job for
    the citizens of Florida than the techniques of
    regression that have been around for decades.
    The other question is how and when
    can human discretion be best combined with
    the dry, cold and unfeeling application of computer rules.