Portable, non-destructive colorimetry and visible reflectance spectroscopy paired with machine learning can classify experimentally heat-treated silcrete from three South African sources

• DOI: 10.1371/journal.pone.0266389
• Publication Date: 2022-04-08T17:29:40Z
• Authors: John K. Murray; Simen Oestmo; Andrew M. Zipkin
• Keywords: Colorimetry; Replicate

The objective of this study was to determine if visible reflectance spectroscopy and quantitative colorimetry represent viable approaches classifying the heat treatment state silcrete. Silcrete is a soil duricrust that has been used as toolstone since at least Middle Stone Age. ancient practice treating silcrete prior knapping considerable interest paleolithic archaeologists because its implications for early modern human complex cognition generally ability manipulate material properties stone specifically. Here, we demonstrate our quantitative, non-invasive, portable approach measuring color, in conjunction with k-Nearest Neighbors "lazy" machine learning, highly promising method detection. Traditional, expert analyst typically rely upon subjective assessments color luster comparison experimental reference collections. This strongly visual can prove quite accurate, but difficult reproduce between different analysts. In work, measured percent spectrum (1018 variables) standardized values (CIEL*a*b*) unheated experimentally heat-treated specimens from three sources South Africa. k-NN classification proved effective both data sets. An important innovation using predicted by model majority replicate observations single specimen predict overall. When voting applied 746 individual study, associated 94 discrete flakes, models yielded 0% test set misclassification rates level.