The Yilgarn Craton is one of the world’s largest fragments of Archaean crust and among the most mineral-rich regions globally. Granites comprise >70 per cent of the surface area and are linked to major mineral systems, including gold (Au), lithium (Li), and rare earth elements (REE). Robust classification and mapping of granite suites are therefore fundamental to understanding craton-scale architecture, early Earth geodynamics, and metallogenic fertility. Traditional geological mapping of the Yilgarn Craton has relied on manual interpretation of geophysical data supplemented by field observations. While effective at local scales, this approach is labour-intensive, subjective, and difficult to scale consistently, particularly in covered terrains. Existing granite classifications are largely based on petrographic, geochemical, geochronological, and isotopic analyses of outcrop samples. Although these studies successfully define granite groups, they are constrained by exposure bias and limited capacity to integrate large, multi-parameter geophysical data sets coherently across the entire craton. Here, we apply machine-learning (ML) techniques to integrate regional gravity, magnetic, and radiometric data sets to classify granites across the entire Yilgarn Craton — at a scale not previously demonstrated. Both unsupervised and supervised ML approaches are applied. Unsupervised clustering identifies domains with similar petrophysical signatures, while supervised models produce a predicted granite classification map calibrated to granite classes derived from geochemical data. Unsupervised clustering identifies coherent craton-scale domains that closely align with recognised terrane boundaries and independently derived Sm-Nd and Lu-Hf isotopic provinces. A sharp contrast in granite clusters is present between the Youanmi and South West terranes. Within the Youanmi Terrane, clustering supports similarities between the obsolete Southern Cross and Murchison domains, supporting disestablishment of the two domains, while also identifying a distinct granitic and textural domain along the western margin of the terrane, suggesting a previously unrecognised internal subdivision. Along the southern segment of the boundary between the Youanmi Terrane and the Eastern Goldfields Superterrane (EGST), a north-trending belt of similar clusters extend further east than currently mapped, implying that a revision of the terrane boundary is required. A supervised ML-derived granite classification identifies a strong craton-scale contrast between the EGST and older terranes of the western Yilgarn Craton. The EGST is dominated by High-Ca, high Sr/Y granites, whereas the Youanmi, South West, and Narryer terranes are dominated by Low-Ca granites. Low-Ca (high-Ti) granites occur almost exclusively within the most isotopically evolved parts of the craton. Our newly identified granitic subdivision within the Youanmi Terrane likely represents different crustal exposure levels related to variations in metamorphic grade, indicating existence of a previously unrecognised major crustal boundary. These ML results have direct implications for mineral exploration. Newly defined terrane boundaries and internal granite domains may represent favourable zones for Au and Li-pegmatite mineralisation. Low-Ca granites are predicted beneath shallow cover in the Yamarna Terrane, opening new exploration search space for potential Li-pegmatite mineralisation.Ultimately, this study demonstrates the power of integrating unsupervised and supervised ML with regional geophysics to resolve crustal architecture and bedrock geology at the craton scale. The approach is transferable to smaller scales, offering significant potential for greenstone belt-scale targeting of Li, Au, and other critical commodities using high-resolution geophysical data, de-risking exploration in underexplored or covered terrains.