MESOSAT is a fault-tolerant ensemble model for the classification of archaeological sites. The intelligent system incorporates three ML models, a majority voter, fault injection mechanisms, and visualisation generation in the form of confusion matrices and heatmaps. MESOSAT outputs worthiness statistics, classed as “Yes” for excavation worthiness, “No” for unworthiness, and “Partial” for possible excavation, supplemented with the training loss, epochs, accuracy, precision, recall, and F1 scores.

A dedicated dataset was created for this project, containing images from over 20 Mesopotamian archaeological sites. The sites included Sippar, Assur, Nippur, Lagash, Borsippa, Ur, Tell Zurghul, Library of Ashurbanipal, Borsippa Ziggurat, Nimrod Ziggurat, Assur Ziggurat, Palace of Sennacherib, Nergal Palace, Hanging Gardens of Babylon, Uruk, Eridu, Tell Harmal, Nuzi, Taq Kisra, Theatre of Babylon and more. The sites belong to various Mesopotamian civilisations, including the Sumerian, Assyrian, Akkadian and Babylonian.