How did prehistoric people choose where to live?A machine-learning case study from Latium vetus

What is it about?

Why did prehistoric communities choose particular places to establish their settlements? This study investigates how environmental factors influenced settlement location in Latium vetus, the region of central Italy that later became the setting for the emergence of Rome. The analysis covers more than 1,300 years, from the Early Bronze Age to the beginning of the Iron Age (c. 2100–725 BCE). I used a machine-learning method called Random Forest to analyse 108 securely dated settlements together with a set of simulated locations where settlements have not been documented. The model considers factors such as topography, elevation, distance from rivers, springs, lakes and the sea, and the availability of different types of agricultural land. Importantly, the aim is not to predict where undiscovered settlements should be located, but to identify which environmental variables were most strongly associated with settlement choices. Explainable AI techniques, particularly SHAP (SHapley Additive exPlanations), make it possible to examine not only which variables were important, but also how their influence changed through time. The results reveal a significant long-term transformation. During the earlier Bronze Age, access to water was a major factor distinguishing settlement locations. From the Middle Bronze Age onwards, topographic characteristics and elevation progressively became more important. By the Final Bronze Age and the subsequent Roma-Colli Albani phases, settlements were increasingly concentrated in distinctive and potentially defensible landforms. The study shows how explainable machine learning can provide a quantitative way of investigating changing settlement strategies while remaining closely connected to archaeological interpretation.

Featured Image

Photo by JR Harris on Unsplash

Photo by JR Harris on Unsplash

Why is it important?

Understanding why settlements were established in particular locations is central to reconstructing how prehistoric communities interacted with, adapted to, and transformed their landscapes. Archaeologists have long recognised the importance of factors such as access to water, agricultural resources and defensible locations, but identifying their relative importance and how this changed through time is difficult using conventional approaches. This study provides a way to address this problem quantitatively. By combining Random Forest modelling with explainable AI, it is possible to move beyond a simple identification of correlations and examine how individual environmental variables contribute to the distinction between settlement locations and the wider landscape. The analysis also reveals non-linear and threshold-like relationships that can be difficult to detect with conventional statistical methods. An important advantage is that environmental variables are expressed in quantitative and reproducible terms—for example, walking time to water sources, elevation, and the extent of different land-use classes. This creates a basis for comparing the relative influence of environmental factors not only through time within Latium vetus, but also across different archaeological regions and settlement systems. The study therefore demonstrates how explainable machine learning can complement archaeological reasoning by providing a transparent and quantitatively comparable framework for testing and refining interpretations of settlement organisation and landscape change.

Perspectives