Application of machine learning on isotopic data from tooth microsections for reconstructing weaning patterns and physiological stress

Abstract

The recent development of measuring stable isotope ratios (δ15N and δ13C) in collagen from tooth microsections provides temporal resolution of dietary changes and has been widely used for the reconstruction of breastfeeding, weaning and physiological stress in archaeological datasets. We applied incremental dentine analysis, measuring the δ15N and δ13C in collagen from first permanent molars of 45 adult individuals from the city of Thessaloniki (4th c. BC - 16th c. AD). We were able to reconstruct the diet of 31 individuals from birth up to the age of seven. To this dataset, we added 20 previously published individuals from the same site and we re-examined the weaning ages with WEAN, an automated application for estimating the weaning age based on the measurements of δ15N. Furthermore, we used the k-means machine learning method to discern clusters of different isotopic patterns in the individual profiles and trace possible signals of physiological stress.Our results show that 45 out of 51 individuals were breastfed but weaned at different ages ranging from one to three years old. Five individuals were breastfed for an even shorter period or were never breastfed. The weaning diet was comprised mostly of animal protein and C3 plants, while the consumption of small fish and/or C4 plants intensified from the Roman period onwards. There were no statistically significant differences between males and females and among chronological periods. Based on the machine learning approach we could identify 9 individuals with probable evidence of physiological stress, which may be linked to maladaptive breastfeeding patterns. Our study reports new data on breastfeeding and weaning utilizing the implementation of computational methods and illustrates the complexity of the infant feeding practices in ancient societies.