Development and Validation of an Internal Prison Classification using Machine Learning Techniques

• In Event: The Role of Risk/Needs Assessment in the Correctional System

Abstract

Internal prison classification systems emerged in the 1980s. Early exemplars included Megargee’s MMPI typology, Quay’s Prison System (AIMS) and others. Van Voorhis (1995) provided an early evaluation of these systems. Internal Classification emerged because of deficiencies of simple security classifications in offering guidance for multiple management and programming decisions for prisoners. A more recent national evaluation identified multiple problems with available internal classifications, concluding that these systems represented a relatively “primitive” state of development(NIC 2002). The present study reports the development and validation of a theory-guided Internal Classification for two state DOC systems. It addresses several design and performance flaws of prior systems by using a theory-guided approach to the measurement domain, more reliable psychometric measures and by including the Risk-Need-Responsivity (RNR) principles into the classification design. Pattern analytic bootstrap methods were used to empirically identify eight robust and interpretable offender categories and cross sample stability tests were applied. The Random Forests (RF) approach reached accurate and reliable automated identification of new cases into the classification system.

Authors

• Tim Brennan, Northpointe, Inc.

• William L. Oliver, Northpointe, Inc.

• William Dieterich, Northpointe, Inc.