Predicting Flake Mass: A View from Machine Learning

Abstract

ABSTRACT Estimating flake mass based on remaining attributes bears an important relationship for the interpretation of lithic assemblages. Previous works have pointed out the relationship between flake attributes and prediction of flake mass. This study builds on previous works by using data from an experimental collection of flakes. Estimated mass was arrived at by generating a multiple linear regression model that combines several predictive variables. Variable selection for model training was carried out by using best subset selection, which evaluates all possible combinations of variables. Evaluation of the model was performed by computing common machine learning statistics along with estimated percentage error. Results make it possible to determine the best variables and estimate their relationships with flake mass. On the other hand, results also show that although the model is slightly biased and performs adequately, it has a limited inferential ability, especially when compared with other methods/indexes employed to estimate reduction.