Preservation of historic masonry buildings requires the precise identification of structural issues, such as foundation settlements, which can compromise stability. Traditional detection methods rely on visual inspections, non-destructive testing, or complex numerical simulations, which can be time-consuming and may not provide real-time data for intervention. To overcome these limitations, this study proposes a machine learning (ML) methodology capable of identifying both the position and severity of foundation settlements by analysing the static and dynamic response of masonry buildings. The methodology involves training a ML algorithm using a dataset derived from a finite element model of the target building. By leveraging data on structural displacements, rotations, modal frequencies, and mode shapes under various settlement scenarios, the model can predict the location and magnitude of foundation displacements while significantly reducing computational costs compared to traditional finite element analyses. In this study, the methodology is tested on a regular isolated stone masonry building, representative of typical Italian residential structures from the 18th and 19th centuries, where planar foundation settlements with varying directions, locations, and magnitudes are applied. The proposed approach enables real-time monitoring and early detection of structural issues, providing a nonintrusive and cost-effective solution for the maintenance of historic structures.