Title: Machine Learning Classification with Experimental Bone Surface Modifications
Speaker: Jordy Orellana Figueroa, ERC STONECULT Project, Tübingen
Date: Tuesday, 30 June 13:00 s.t.
Venue: Zoom, https://zoom.us/j/91423207550, Meeting ID: 914 2320 7550
The identification of actors and effectors of bone surface modifications (BSMs) in the archaeological record is often a source of contention. Many debates centre on whether a given BSM was made by stone tool-using hominins, carnivore predation, and/or trampling. Much research has similarly attempted to accurately assess the agent of BSMs based on the morphology and configuration of marks, as this could have important implications for our understanding of human evolution (e.g. McPherron et al. 2010; Sahle et al. 2017).
I present here the initial results of a machine learning model trained to classify a dataset of images generated from experimental BSM marks. The model classifies images of BSM marks into two categories based on the agent of modification: lithic and non-lithic. The dataset was kindly provided by Dr Yonatan Sahle, and consisted of a small number of surface scans and 3D models of BSMs from experimental butchery using stone tools and an actualistic crocodile (Crocodylus niloticus) feeding experiment. All data were generated from high-resolution casts that were scanned using a confocal microscope. The model, using an artificial neural network, obtained a mean classification accuracy of 92.60% across 30 runs. Additional tests were performed with various model parameters to evaluate the model’s robusticity, obtaining comparable results. Even so, additional training data will be required so that blind tests can be conducted and a more definitive answer on the strength of this approach can be given.
This presentation will outline the methodology used in this project, provide a detailed interpretation of the results, and discuss future steps of this research.