LAUSR

Abstract Extensive efforts have focused on preventing or reducing the prevalence of sexual assault and sexual homicide of children and adults. Offender, victim, offence, and environmental factors have been associated with sexual assault injury and death more broadly, however, little to no prior research has investigated the role of these factors in explaining lethality using machine learning techniques. Using a database of 624 cases of sexual assault containing offender, victim and situational variables, a supervised machine learning technique is utilized to identify characteristics and situations that are most predictive of victim death during sexual assault. Results identified proximal predictors of lethality during sexual assault, including fighting between victim and offender within the 48 h prior to the assault and offender weapon possession. Greater victim-offender intimacy reduced the likelihood of assault lethality. One offence characteristic (forcing victim to perform sex act) also predicted reduced lethality. Findings demonstrate the potential use of machine learning techniques to identify the most salient predictors of lethality during sexual assault and to produce knowledge that will contribute to the development of lethality assessment for sexual assault.