Investigating predictors of juvenile traditional and/or cyber offense using machine learning by constructing a decision support system

Abstract

The present study aims to examine whether criminogenic risk factors can be applied to explain different types of juvenile offenses involving traditional and/or cyber offenses, and explore their common and unique patterns presented among juvenile offenders. To achieve the goals, this study employs machine learning (ML) techniques to construct a decision support system that predicts different types of juvenile offenses (i.e., non-offense, hacking only, traditional offense only, and both offenses) by risk factors rooted in a variety of criminological theories. This study is based on the data from the Second International Self-Report of Delinquency Study. The results demonstrate the generalizability of mainstream criminological theories to juvenile hacking and dual offenses involving both traditional offense and hacking. ML predictive models can successfully distinguish between different types of juvenile offenders and identify the most influential risk factors (e.g., gender, digital piracy, substance use, victimization, and parental supervision). The relative importance of risk factors provides valuable information to decision-makers and stakeholders in the juvenile justice system for developing more effective risk assessments and early intervention programs targeting different types of juvenile offenders.