Author, as appears in the article.: Guillermo Bustos-Pérez, Javier Baena Preysler
Department: Història i Història de l'Art
URV's Author/s: Bustos-Pérez, Guillermo
Abstract: Predicting original flake mass is a major goal of lithic analysis. Predicting original flake mass allows for re- searchers to make estimations of remaining mass, lost mass, and other features. All these measures relate to the organization of lithic technology by past societies. The present work tests three different models to predict log of flake mass: multiple linear regression, random forest regression, and artificial neural networks (ANN). Estima- tions of flake mass were performed using the remaining features of flakes from an experimental assemblage. This assemblage was obtained by the expansion of a previous dataset through the inclusion of bigger flakes, allowing the analysis to account for the effects of sample size and value distribution. Correlation results show a large/ strong relation between predictions and real outcome (r2 = 0.78 in the best case). Comparison of the models affords insights into variable importance for predicting flake mass. Results show that (for the present dataset) multiple linear regression still stands as the best method for predicting log of flake weight. Additionally, transformation of predicted values from the multiple linear regression and true values to the linear scale re- inforces the linear correlation above the 0.8 threshold.
licence for use: https://creativecommons.org/licenses/by/3.0/es/
Author identifier: 0000-0002-1089-818X
Papper version: info:eu-repo/semantics/publishedVersion
Link to the original source: https://www.sciencedirect.com/science/article/pii/S2352409X22003613
APA: Bustos-Pérez, G., & Baena Preysler, J. (2022). Multiple approaches to predicting flake mass. Journal of Archaeological Science: Reports, 46, 103698. https://doi.org/10.1016/j.jasrep.2022.103698
Licence document URL: https://creativecommons.org/licenses/by/3.0/es/
Article's DOI: 10.1016/j.jasrep.2022.103698
Journal publication year: 2022
Publication Type: info:eu-repo/semantics/article
Repositori URV