Understanding sexual homicide in Korea using machine learning algorithms.

The current study was conducted to confirm the characteristics in sexual homicide and to explore variables that effectively differentiate sexual homicide and nonsexual homicide. Further, newer methods that have received attention in criminology, such as the machine learning method, were used to explore the ideal algorithm for classifying sexual homicide and patterns for sexual homicide in Korea. To do this, 542 homicide cases were analyzed utilizing eight algorithms, and the classification performance of each algorithm was analyzed along with the importance of variables. The results of the analysis revealed that the Naive Bayes, K-Nearest Neighbors, and RF algorithms demonstrate good classification accuracy, and generally, factors such as relationships, marriage, planning, personal weapons, and overkill were identified as crucial variables that distinguish sexual homicide in Korea. In addition, the crime scene information of the crime occurring in the dark (at night) and body disposal were found to have high importance. The current study proposes ways to enhance the efficacy of crime investigation and advance the research on sexual homicides in Korea through a more scientific understanding of sexual homicide that has not been thoroughly explored domestically.

Crime Scene Behaviors and Characteristics of Offenders with Mental Illness: A Latent Class Analysis.

The current study aimed to identify distinct types of crime scene behaviors based on the criminal planning and motivation of offenders with mental illness in South Korea. Furthermore, our study examined the relationships between the identified types of crime scene behaviors in terms of the offenders' sociodemographic characteristics, modus operandi, and types of mental illness. Utilizing latent class analysis, the associations between crime scene behavior types and offender characteristics such as demographic factors, crime scene actions, and criminal information were empirically investigated. In particular, based on a sample obtained from a national police database of offenses committed between 2006 and 2014, four offense groups were identified: (i) instrumental-planned, (ii) instrumental-unplanned, (iii) expressive-unplanned, and (iv) hybrid. Additionally, significant relationships were found between offense styles and offender characteristics as well as criminal backgrounds. The findings suggest that mental disorders influence the types of actions exhibited by offenders during the commission of their crime. The results are discussed in terms of their theoretical and practical utility to criminal investigation.