Some time between 1750 and 1650 BCE in the Egyptian city of Elephantine, people living a house left a mess. Malleson & Srour used machine learning to analyse the mess, and found that it wasn’t entirely random. A study of of 208,493 plant items from 123 archaeological contexts showed that some of the site were locations for “waste” disposal. Other deposits revealed places flax was processed and the locations of fireplaces.
Malleson & Srour tackled the plant remains in house 169 on Elephantine Island. They had a massive number of charred and desiccated plant remains to analyse. Most of these remains would be expected to be the result of intentional waste disposal, but the archaeobotanists wanted to know if something could be inferred about the house from the distribution of plant remains.
The scientists used two main approaches using machine learning: clustering, which groups similar items together, and evolutionary decision trees, which help identify important factors influencing these groups. This allowed them to analyse every archaeological context, rather than one at a time.
The analysis identified barley chaff, fig remains and leftovers from grain processing throughout the house, but did also identify cleaner and dirtier areas. Some rooms showed signs of specific activities – for example, one room had lots of flax remains, suggesting it might have been used for processing this important plant used to make linen. Another area had higher concentrations of grain, hinting that it might have been used for storage.
Malleson & Srour note that machine learning helped make sense of what appeared to be a homogenous mess. “It was only via the ML (clustering and prediction trees) that broad differences in the assemblage were revealed; “clean” vs. “dirty” deposit types.”
Malleson, C., & Srour, F. J. (2024). “It’s all just barley and figs!” Identifying patterns of plant waste accumulation in House 169, Elephantine Island, Egypt (1750–1650 BC) using machine learning. Vegetation History and Archaeobotany. https://doi.org/10.1007/s00334-024-01010-x
ReadCube: https://rdcu.be/dTv0l
