A proposal for cut marks classification using machine learning: Serrated vs. non-serrated, single vs. double-beveled knives.

In tool mark identification, there is still a lack of characteristics and methodologies standardization used to analyze and describe sharp force trauma marks on skeletal remains. This study presents a classification method for cut marks on human bones, providing an applicable methodology for their examination and the relevant terminology for describing cases of sharp force trauma. A total of 350 cut marks were produced by stabbing pig ribs (Sus scrofa) with seven knives. The samples were analyzed under a stereomicroscope with a tangential light source. Through the analysis of cut marks, eleven traits were identified as significantly associated with the type of knife used. These traits included the general morphology of the kerf shape, the entrance and exit cross-profile shapes, the location of the rising on the entrance and exit cross-profile, the presence or absence of feathering, the presence or absence of shards and the location and the general morphology of the mounding. Binary logistic regression models were later trained and tested using nine out of the eleven traits. The first model categorized the cut mark as either produced by a serrated or non-serrated blade, while the second, as either produced by a single- or double-beveled blade. Classification scores of those models ranged between 63%-85% for the serration class and 63%-89% for the blade bevel class. This study proposes a new set of traits and the use of machine learning models to standardize and facilitate the analysis of stab wounds.