Natalia Abellán , Enrique Baquedano , Manuel Domínguez-Rodrigo
{"title":"High-accuracy in the classification of butchery cut marks and crocodile tooth marks using machine learning methods and computer vision algorithms","authors":"Natalia Abellán , Enrique Baquedano , Manuel Domínguez-Rodrigo","doi":"10.1016/j.geobios.2022.07.001","DOIUrl":null,"url":null,"abstract":"<div><p>Some researchers using traditional taphonomic criteria (groove shape and presence/absence of microstriations) have cast some doubts about the potential equifinality presented by crocodile tooth marks and stone tool butchery cut marks. Other researchers have argued that multivariate methods can efficiently separate both types of marks. Differentiating both taphonomic agents is crucial for determining the earliest evidence of carcass processing by hominins. Here, we use an updated machine learning approach (discarding artificially bootstrapping the original imbalanced samples) to show that microscopic features shaped as categorical variables, corresponding to intrinsic properties of mark structure, can accurately discriminate both types of bone modifications. We also implement new deep-learning methods that objectively achieve the highest accuracy in differentiating cut marks from crocodile tooth scores (99% of testing sets). The present study shows that there are precise ways of differentiating both taphonomic agents, and this invites taphonomists to apply them to controversial paleontological and archaeological specimens.</p></div>","PeriodicalId":55116,"journal":{"name":"Geobios","volume":null,"pages":null},"PeriodicalIF":1.6000,"publicationDate":"2022-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"3","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Geobios","FirstCategoryId":"89","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0016699522000377","RegionNum":4,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"PALEONTOLOGY","Score":null,"Total":0}
引用次数: 3
Abstract
Some researchers using traditional taphonomic criteria (groove shape and presence/absence of microstriations) have cast some doubts about the potential equifinality presented by crocodile tooth marks and stone tool butchery cut marks. Other researchers have argued that multivariate methods can efficiently separate both types of marks. Differentiating both taphonomic agents is crucial for determining the earliest evidence of carcass processing by hominins. Here, we use an updated machine learning approach (discarding artificially bootstrapping the original imbalanced samples) to show that microscopic features shaped as categorical variables, corresponding to intrinsic properties of mark structure, can accurately discriminate both types of bone modifications. We also implement new deep-learning methods that objectively achieve the highest accuracy in differentiating cut marks from crocodile tooth scores (99% of testing sets). The present study shows that there are precise ways of differentiating both taphonomic agents, and this invites taphonomists to apply them to controversial paleontological and archaeological specimens.
期刊介绍:
Geobios publishes bimonthly in English original peer-reviewed articles of international interest in any area of paleontology, paleobiology, paleoecology, paleobiogeography, (bio)stratigraphy and biogeochemistry. All taxonomic groups are treated, including microfossils, invertebrates, plants, vertebrates and ichnofossils.
Geobios welcomes descriptive papers based on original material (e.g. large Systematic Paleontology works), as well as more analytically and/or methodologically oriented papers, provided they offer strong and significant biochronological/biostratigraphical, paleobiogeographical, paleobiological and/or phylogenetic new insights and perspectices. A high priority level is given to synchronic and/or diachronic studies based on multi- or inter-disciplinary approaches mixing various fields of Earth and Life Sciences. Works based on extant data are also considered, provided they offer significant insights into geological-time studies.