Age Estimation with Decision Trees: Testing the Relevance of 94 Aging Indicators on the William M. Bass Donated Collection

Anthropologists have been estimating ages-at-death of skeletons for a long time. A variety of different age indicators has been studied and age estimation methods have been developed in an attempt to standardize the process. Even with all the work that has gone into developing age estimation methods, age estimation of mature skeletons is still very imprecise. This research investigates various age indicator definitions and their performance on an elderly skeletal sample. Using 176 individuals from the William M. Bass Donated Collection curated in the Department of Anthropology at the University of Tennessee, Knoxville, data were collected on age indicators gathered from fifteen age estimation methods. Ninety-four variables were tested with various decision trees to show patterns among the variables. Regression equations were built using the same variables as the decision trees, and the performance between the two methodologies were compared. The decision trees performed slightly better, with a mean absolute error of prediction of around five years. Variable occurrence was tabulated across various decision tree models. The most common variables are pit shape of the sternal rib end morphology and the phase of the auricular phase. These two variables, along with others commonly selected, present best candidates for building an age estimation method that pertains to older populations.