Laura Vodden
Stone artefacts comprise a large proportion of objects recovered from archaeological sites, owing to their durability and resistance to deterioration (Clarkson and O’Connor, 2007). They are valuable to archaeologists, because they provide an excellent record of human behaviour throughout prehistory, and can reveal information about diet, creativity, and past human interactions with the environment (Clarkson and O’Connor, 2007; Marwick, 2008). The purpose of this report is to investigate whether machine learning algorithms can use archaeological data to classify a stone artefact assemblage into two broad categories. The report makes use of observational data relating to artefacts excavated from a rock shelter at Ban Rai, Northwest Thailand, between 2001 and 2007. In performing the investigation, three machine learning algorithms were applied to the data: A Naïve-Bayes Classifier first assessed the accuracy of predictions, a Logistic Regression then assessed the variables most likely to affect classification either way, and a K-Means Cluster Analysis was used to visualise the data, to see if the parameters for each artefact category were distinct enough to form two clusters. The report finds that machine learning algorithms can be used to classify the Ban Rai stone artefacts into two categories. This has implications for archaeological investigation, because it suggests that machine learning algorithms can be applied to other archaeological assemblages, and perhaps can even further differentiate between sub-categories of artefacts, leading to new and exciting discoveries.