In This Chapter

An Alternative Scientific Paradigm for Criminological Risk Assessment

Closed or Open Systems, or Both?

Authored by: Tim Brennan

Handbook on Risk and Need Assessment

Print publication date:  October  2016
Online publication date:  November  2016

Print ISBN: 9781138927766
eBook ISBN: 9781315682327
Adobe ISBN:


 Download Chapter



In a recent debate in Criminology and Public Policy (CPP) Richard Berk and Justin Bleich (2013) in “Statistical Procedures for Forecasting Criminal Behavior” compared the predictive performance of two Machine Learning (ML) procedures (Random Forests and Gradient Boosting) against the currently dominant multiple regression method (MR). They concluded that the ML methods made fewer errors, particularly with more complex data characterized by non-linear boundaries and other complex data structures. However, Berk and Bleich noted an unusual contradiction in the available comparative studies. Specifically, the predictive accuracy of ML methods versus MR was higher when conducted by computer scientists and machine learning laboratories than when such comparisons were conducted by criminal justice researchers on criminal justice data. Some criminal justice researchers, on the basis of these studies, concluded that ML methods have no clear advantage and that there is little need to adopt or explore these new methods.

Search for more...
Back to top

Use of cookies on this website

We are using cookies to provide statistics that help us give you the best experience of our site. You can find out more in our Privacy Policy. By continuing to use the site you are agreeing to our use of cookies.