Can machine learning help criminologists to characterize a serial killer? A retrospective pilot study on an Italian sample.

Machine learning is a subset of the analysis methods known as 'artificial intelligence', which can automatically construct models from a great amount of data. Limitation of the method is that an incorrect attribution of a parameter to one class rather than another can lead to misleading results. In the case of small datasets, adding or removing a case can change the results substantially. Serial killing is a rare phenomenon in which one (rarely more) person kills two or more people to satisfy his own psychological gratification. Many other reasons for serial killing include anger, financial gain and sexual impulse. Although a serial killer differs from other kinds of multiple homicide attenders (such as mass murderer, spree killer, or contract killer) there exist conceptual overlaps between them. The aim of this study is to perform a data driven analysis using machine learning of the Italian serial killers of the 20th century. A database was filled searching on the internet for all homicide perpetrators classified as “serial killer” (one record for each homicide perpetrator). For each biographical data, behavioral characteristics and judicial information were also searched. Unsupervised data processing methods [clustering and dimensionality reduction] were then used and sub-groups were obtained. For this purpose, “Orange3”' [https://orangedatamining.com/] was used, which is an open source machine learning and data visualization software that builds data analysis workflows in a visual way, with a large and diverse set of tools. Despite the low number of cases processed, the relatively few parameters examined and the lack of some information on unsolved cases, the results obtained seem to be promising.