Classification of Bovidae fossils from Gladysvale, South Africa using elastic shape analysis

Teeth from the Family Bovidae that are associated with our early humans ancestors are important for reconstructing paleoenvironments. However, age, degree of attrition, and taphonomic factors often make fossil identification difficult. A recent technique for classifying these teeth uses the size-and-shape of the occlusal surface as a summary of the surface, deriving features from this, and then using these features in machine learning classification algorithms. Bovid teeth have previously been classified using this method with features derived from coefficients of elliptical Fourier analysis (EFA). This study assesses the utility of using other shape representations for feature generation, specifically elastics shape analysis. Features were derived using this frame work for both shape only and size-and-shape (i.e. size is not considered a nuisance parameter), and those features were used as input for machine learning algorithms. We demonstrate that features derived elastic shape analysis generally outperform features derived from EFA in terms of cross validation classification accuracy. Finally, an application of the classification methods studied here was applied to fossils recovered from the deroofed Gladysvale External deposit (GVED), Gauteng Province, South Africa. Previous analyses of GVED identified a group of bovids as medium sized alcelaphines (Lacruz et al., 2002). Specifically, this study reclassified 32 unbroken, medium sized alcelaphines looking at shape and size-and-shape. The reclassifications increased the number of individuals and diversity of bovids recovered from the site. The results were used to generate a more precise paleoenvironmental reconstruction.