How Machine Learning Reveals Ancient Settlement Shifts in Italy

A new study looks at how ancient settlements moved across central Italy. It covers the period from the early Bronze Age to the early Iron Age, about 2100 to 725 BCE.

The team used a Random Forest algorithm, a type of machine learning, together with explainable AI tools. They examined 108 sites that have solid dates. They also added pseudo‑absence points to fill gaps in the data. SHAP values helped show which factors mattered most.

At the start, the model shows that water sources guided where people built homes. Later, as the Bronze Age progressed, the shape of the land and its height became the key. Settlements moved to spots that were easier to defend and looked different. In the final Bronze and RMCA phases, the pattern became very regular, like canals. The later RMCAIII phase shows a landscape that is more layered, with many functions, even though politics stayed steady.

The study notes that using fake absence points and uneven site numbers can weaken the model. Still, the approach proves that explainable machine learning can give a clear, data‑driven picture of how ancient lands changed. It helps archaeologists test old ideas with modern tools.