Open AccessAssessing the Risk of Repeat Victimization Using Structured and Unstructured Police Information
Author(s) -
Roos Geurts,
Niels Raaijmakers,
Marc J. M. H. Delsing,
Toine Spapens,
Jacqueline A. M. Wientjes,
Dick L. Willems,
Ron H. J. Scholte
Publication year - 2021
Publication title -
crime and delinquency/crime and delinquency
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.431
H-Index - 73
eISSN - 1552-387X
pISSN - 0011-1287
DOI - 10.1177/00111287211047533
Subject(s) - unstructured data , police department , directive , vulnerability (computing) , decision tree , sample (material) , computer science , psychology , computer security , data mining , criminology , big data , chemistry , chromatography , programming language
Following the EU Victim Directive, Dutch police officers are obliged to assess a victim’s vulnerability to repeat victimization. This study explored the utility of unstructured police information for the prediction of repeat victimization, as well as its incremental value over and above structured police information. Police records over a period of 6 years were retrieved for a sample of 116,680 victims. Unstructured information was transformed into numeric features using count-vector and TF/IDF methods. Classification models were built using decision tree and random forest models. AUC values indicate that a combination of structured and unstructured police information could be used to correctly classify a majority of repeat and non-repeat victims.