The Application of Machine Learning to a General Risk–Need Assessment Instrument in the Prediction of Criminal Recidivism

Published date01 April 2021
AuthorMahshid Atapour,Raymond J. Spiteri,Keira C. Stockdale,Daniel Anvari,J. Stephen wormith,Mehdi Ghasemi
DOI10.1177/0093854820969753
Date01 April 2021
Subject MatterArticles
CRIMINAL JUSTICE AND BEHAVIOR, 2021, Vol. 48, No. 4, April 2021, 518 –538.
DOI: https://doi.org/10.1177/0093854820969753
Article reuse guidelines: sagepub.com/journals-permissions
© 2020 International Association for Correctional and Forensic Psychology
518
THE APPLICATION OF MACHINE LEARNING TO
A GENERAL RISK–NEED ASSESSMENT
INSTRUMENT IN THE PREDICTION OF
CRIMINAL RECIDIVISM
MEHDI GHASEMI
University of Saskatchewan
DANIEL ANVARI
Kwantlen Polytechnic University
MAHSHID ATAPOUR
Capilano University
J. STEPHEN WORMITH
KEIRA C. STOCKDALE
Saskatoon Police Service
University of Saskatchewan
RAYMOND J. SPITERI
University of Saskatchewan
The Level of Service/Case Management Inventory (LS/CMI) is one of the most frequently used tools to assess criminogenic
risk–need in justice-involved individuals. Meta-analytic research demonstrates strong predictive accuracy for various recidi-
vism outcomes. In this exploratory study, we applied machine learning (ML) algorithms (decision trees, random forests, and
support vector machines) to a data set with nearly 100,000 LS/CMI administrations to provincial corrections clientele in
Ontario, Canada, and approximately 3 years follow-up. The overall accuracies and areas under the receiver operating char-
acteristic curve (AUCs) were comparable, although ML outperformed LS/CMI in terms of predictive accuracy for the middle
scores where it is hardest to predict the recidivism outcome. Moreover, ML improved the AUCs for individual scores to near
0.60, from 0.50 for the LS/CMI, indicating that ML also improves the ability to rank individuals according to their probabil-
ity of recidivating. Potential considerations, applications, and future directions are discussed.
Keywords: LS/CMI; risk–need assessment; predictive accuracy; machine learning
Although efforts to predict criminal recidivism date back 90 years (Burgess, 1928), the
last two decades have witnessed an explosion in the use of risk-assessment tools in
criminal justice systems around the world. These tools vary dramatically in their length,
AUTHORS’ NOTE: The views expressed are solely those of the authors and do not necessarily reflect those
of the Saskatoon Police Service. In addition, we wish to acknowledge support from the Ontario Ministry of
Community Safety and Correctional Services, the Centre for Forensic Behavioural Science and Justice Studies
at the University of Saskatchewan, the Saskatchewan Police Predictive Analytics Lab, Mitacs, and the Natural
Sciences and Engineering Research Council of Canada. Correspondence concerning this article should be
addressed to Raymond J. Spiteri, Department of Computer Science, University of Saskatchewan, 176
Thorvaldson Building, 110 Science Place, Saskatoon, Saskatchewan, Canada S7N 5C9; e-mail: spiteri@cs.
usask.ca.
969753CJBXXX10.1177/0093854820969753Criminal Justice and BehaviorGhasemi et al. / Application of ML to Risk–Need Assessment
research-article2020
Ghasemi et al. / APPLICATION OF ML TO RISK–NEED ASSESSMENT 519
scope, design, and method of calculating or appraising risk. They also vary in the type of
forensic clientele for whom they are designed, the type of outcome they are meant to predict
(e.g., types of recidivism), and the context in which they are applied (Andrews et al., 2006).
Yet, they also tend to have some common characteristics. For example, most risk-assess-
ment tools capture data about an individual’s criminal history, a so-called static or historical
factor and perhaps the most well-established risk factor of subsequent criminal behavior.
Another characteristic that binds all forensic risk-assessment instruments is that they are
ultimately intended to promote public safety by identifying individuals who are most likely
to reoffend. It is then the responsibility of the criminal justice system (police, courts, cor-
rectional agencies, and community organizations) to use the results of forensic risk assess-
ments to employ the appropriate means at their disposal to reduce or prevent further criminal
behavior.
THE LEVEL OF SERVICE (LS) FAMILY OF RISK-ASSESSMENT TOOLS
The Level of Service/Case Management Inventory (LS/CMI; Andrews et al., 2004) is the
latest version of a forensic risk–need assessment measure from a family of tools known as
the LS scales. Versions of the LS scales have been used worldwide since the early 1990s,
with increasing popularity over the last decade. For instance, by 2010, more than one mil-
lion administrations were officially registered with the test publisher in a single year
(Wormith, 2011). The popularity of the LS scales may be attributed to several important
characteristics. First, unlike strictly actuarial measures, the LS scales were developed from
well-established criminological and psychological theories (e.g., differential association
theory, social learning theory), including a general personality and cognitive social learning
theory of criminal behavior (e.g., Andrews & Bonta, 1994). Second, the LS scales have a
rich tradition of research supporting its content and use in practical ways for correctional
practitioners (Gendreau et al., 1996). This includes numerous validation studies and meta-
analyses (e.g., Olver et al., 2014). Third, the LS scales have been found to have general
applicability across many forensic populations. This includes adults and youth in custody or
on community supervision, male and female populations, and various ancestral/ethnic
backgrounds and cultures on diverse measures of recidivism, ranging from technical viola-
tions to criminal charges and convictions (e.g., Olver et al., 2009; Smith et al., 2009; Wilson
& Gutierrez, 2014). Fourth, the LS scales have multiple applications in corrections. This
includes not only the prediction of criminal recidivism but also the planning and delivery of
forensic services and case management practices to prevent recidivism (e.g., Luong &
Wormith, 2011), an attribute made possible because the scale includes dynamic risk factors,
also known as criminogenic needs, as well as static risk factors, hence its status as a risk–
need scale. Fifth, the LS scales strike a balance between comprehensiveness and simplicity.
Ratings in applied settings require a skilled interview of forensic clientele, yet items are
scored in a dichotomous (0–1) fashion and then summed. As such, it can easily be scored
manually by a trained assessor.
A pilot version of the LS/CMI called the Level of Service Inventory–Ontario Revision
(LSI-OR; Andrews et al., 1995) was introduced in Ontario, Canada, in 1995, and remains in
use throughout this provincial jurisdiction. More than 20,000 administrations of this version
are applied to forensic clientele in Ontario annually. For simplicity, in this study, we use the
more generally known and widely used name for this version of the tool, the LS/CMI.

To continue reading

Request your trial

VLEX uses login cookies to provide you with a better browsing experience. If you click on 'Accept' or continue browsing this site we consider that you accept our cookie policy. ACCEPT