The Emergence of Machine Learning Techniques in Criminology

• DOI: http://doi.org/10.1111/1745-9133.12055
• Author: William L. Oliver,Tim Brennan
• Date: 01 August 2013
• Published date: 01 August 2013

POLICY ESSAY

FORECASTING CRIMINAL BEHAVIOR

The Emergence of Machine Learning

Techniques in Criminology

Implications of Complexity in our Data and in Research

Questions

Tim Brennan

Northpointe, Inc., Georgia State University

William L. Oliver

Northpointe, Inc.

We view Berk and Bleich’s (2013, this issue) article as potentially an impor-

tant contribution to both criminology and practical criminal justice decision

making. Although it is focused on new machine learning (ML) methods for

forecasting, the article ramifies into several general problems that challenge the dominant,

widely used standardized methods in criminology. The initial challenge raised by Berk

and Bleich concerns the complexity of the data underlying many substantive and theo-

retical issues in criminology and criminal justice (e.g., multidimensionality, nonlinearity,

complex interactions, feedback loops, and multimodality). Second, this challenge leads to

the question of whether our standard parametric methods for predictive forecasting (e.g.,

logistic regression) are mismatched to the complexity and nonlinear decision boundaries

found in many research and practical situations. Third, although ML methods seem to have

substantial forecasting advantages in dealing with complex data, several current evaluative

studies in criminal justice have suggested that ML methods have little predictive advantage

over our standard parametric forecasting methods. Berk and Bleich explore reasons for

these conclusions and conduct new comparative evaluations between two ML forecasting

methods (random forests [RF] and stochastic gradient boosting) compared against logistic

regression. In this policy essay, we examine some implications of the complexity issue for

criminal justice method, theory, and practice, and we elaborate on several key issues raised

by Berk and Bleich.

Direct correspondence to Tim Brennan, Northpointe Inc., 211 Old Town Road, Simpsonville, South Carolina

29681 (e-mail: tbrennan38@earthlink.net).

DOI:10.1111/1745-9133.12055 C2013 American Society of Criminology 551

Criminology & Public Policy rVolume 12 rIssue 3