Skip to main content

Microdebitage Soil Analysis of Tzikin Tzakan Stoneknappers (DSI-SRP)

Posted by on Sunday, August 15, 2021 in Arts and Humanities, College of Arts and Science, Completed Research, DSI-SRP, DSI-Supported Research, Social and Behavioral Sciences.

This DSI-SRP fellowship funded Amy Rieth to work in the laboratory of Professor Markus Eberl in the Department of Anthropology during the summer of 2021. Amy is a junior with majors in Anthropology and English.

The project funded by this fellowship investigated the effectiveness of two methods of computer-based soil particle sorting methods. This investigation aimed at streamlining the identification of microdebitage particles within soil samples collected at various anthropological sites. Microdebitage refers to microscopic pieces of stone shrapnel produced by the stone knapping technique used to make weaponry in a litany of historical societies (note: samples used in this study come from Mayan excavation sites in Guatemala, specifically the Nacimiento and Tamarindito sites). In the past, anthropologists have had to sort soil particles and identify microdebitage pieces manually. In recent years, machine learning and deep learning models have been applied to this task in order to improve the reliability and efficiency of soil sample analysis. This project compared these two forms of modeling regarding their accuracy in identifying the particle class of various soil samples. Machine learning techniques implemented randomForest and xgboost models using 48 variables collected on 198600 particles by the PartAn3D analyzer. Deep learning modeling was completed using an Image Classification transformer applied to ~800 images collected by the imaging function of the PartAn3D analyzer. Initial two-class comparisons between machine and deep learning modeling revealed similar effectiveness in particle identification. However, the inclusion of chert and sand particle classes displayed an increase in accuracy in machine learning model identification, while deep learning image classification decreased in accuracy and increased in particle misidentification. As such, at this stage of research, the machine learning method proves most useful in the computer-based identification of soil particles.

In addition to receiving support through a DSI-SRP fellowship, this project was supported and facilitated by the DSI Data Science Team through their regular summer workshops and demo sessions.

Tags: , , , ,