Are Those Teenagers Really Up to No Good? Developing a Predictive Model of Juvenile Crime



Journal Title

Journal ISSN

Volume Title



Juveniles in the U.S. (people younger than 18) are believed to be responsible for most crime and often make up a majority of police contacts. However, for most crime that occurs, we do not know who or how old the offender is, and therefore are left to estimate exactly how much crime is actually committed by juveniles, such as self reports or arrests. Police agencies often use crime data to discern “hot spots” of crime, and may assume those crime clusters are primarily committed by juveniles if they occur after or near a school. Yet, such assumptions (and actions based on those assumptions) could be imprecise and inaccurate. Indeed, in the majority of police contacts with young people no crimes or illegal items are discovered. This points to inconsistencies in both knowledge and policy for juvenile offending. To better understand juvenile crime and inform youth prevention policies and practices, this dissertation focuses on creating a predictive model to determine the probability of whether a reported crime was committed by a juvenile. The predictive model usees characteristics of the offense, spatial and temporal elements and other factors to predict whether an offense was committed by a juvenile. Having a more accurate prediction model of juvenile offending could lead to more precisely targeted prevention initiatives, more cost-effective use of police and community resources, and more effective crime control or prevention of juvenile crime. The development and validity of the model are tested using police data from a police department located in a large metropolitan area in the mid-Atlantic United States. The most successful model can predict juvenile status at about 90% accuracy, and findings indicate potential for machine learning to be used for research and understanding unknown patterns of juvenile crime. Practical implications and limitations are also discussed.



Delinquency, Evidence based policy, Juvenile crime, Machine learning, Predictive policing, Random forest