Profiling Serial Killers Using Multiple Supervised Machine Learning Approaches

Abstract

Criminal profiling has gained a lot of recognition over the years. Profiling is done by experts who use information from a crime scene, to create a serial killer profile. Such a profile consists of serial killer attributes and can include: the gender, race and possible previous activities of the killer. The paper proposes a framework that combines multiple wellknows supervised machine learning techniques to create such a profile. The majority of the proposed approaches obtained a balanced accuracy over 72%, and a predictive accuracy over 80%. The proposed approaches also performed well on a set of other databases, including a single-victim homicide database where it reached a balanced accuracy over 72% and a predictive accuracy over 77%