Bayesian networks for sex-related homicides: structure learning and prediction

Abstract

Sex-related homicides tend to arouse wide media coverage and thus raise the urgency to find the responsible offender. However, due to the low frequency of such crimes, domain knowledge lacks completeness. We have therefore accumulated a large data-set and apply several structural learning algorithms to the data in order to combine their results into a single general graphic model. The graphical model broadly presents a distinction between an offender and a situation-driven crime. A situation-driven crime may be characterised by, amongst others, an offender lacking preparation and typically attacking a known victim in familiar surroundings. On the other hand, offender-driven crimes may be identified by the high level of forensic awareness demonstrated by the offender and the sophisticated measures applied to control the victim. The prediction performance of the graphical model is evaluated via a model averaging approach on the outcome variable offender's age. The combined graph undercuts the error rate of the single algorithms and an appropriate threshold results in an error rate of less than 10%, which describes a promising level for an actual implementation by the police.