In this study, we examine the influence of the mantle and large-scale tectonics on the global mid-ocean ridge (MOR) system. Using solely seismically-inferred upper mantle temperatures below the melting zone (260-600 km) and an interpretable machine learning model (Random Forest and Principal Component Analysis), we predict, with up to 90\% accuracy, the basin of origin of ridge segments without any prior geographic information. Two features provide $>$50\% of the discriminative power: the temperature difference between the mid-layer (340-500 km) and other depths, and the depth-averaged temperature. Our result implies that the large-scale geophysical and geochemical differences observed along the MOR system are reflective, not primarily of shallow processes associated with melting, but of long-term tectonic and convective processes in the mantle that determine the present-day upper mantle temperature structure.