Distinguishing Discoid and Centripetal Levallois methods through machine learning

Area of Science:

• Archaeology
• Experimental Archaeology
• Paleolithic Technology

Background:

• The Discoid and Centripetal Levallois methods are key lithic reduction strategies in the Paleolithic.
• Distinguishing between these methods based on artifact morphology has been a long-standing challenge in lithic analysis.
• Traditional typological and metrical analyses have limitations in clearly differentiating between these reduction techniques.

Purpose of the Study:

• To apply Machine Learning (ML) algorithms to quantitatively analyze and differentiate between Discoid and Centripetal Levallois knapping methods.
• To identify key typometric variables that best discriminate between the products of these two lithic technologies.
• To introduce a novel, data-driven approach to complement traditional methods in the study of lithic reduction strategies.

Main Methods:

• Experimentally knapped flint flakes were produced using both Discoid and Centripetal Levallois techniques.
• Multiple typometric variables were measured from each flake.
• Seven different Machine Learning algorithms were employed to analyze the measured parameters and classify the flakes.

Main Results:

• Statistically significant differences were identified between the products of the Discoid and Centripetal Levallois methods.
• The ML models achieved an accuracy greater than 80% in differentiating between the two knapping methods.
• Key discriminating variables included maximum length, width at 25%, 50%, and 75% of flake length, platform angles, maximum width, and dorsal scar count.

Conclusions:

• Machine Learning provides a powerful and accurate tool for distinguishing between Discoid and Centripetal Levallois lithic technologies.
• The study demonstrates the utility of multivariate ML methods in advancing lithic research.
• This approach offers new quantitative data and a robust methodology for addressing traditional debates in Paleolithic archaeology.