Enhanced nuclear mass predictions using anisotropic kernel ridge regression incorporating multiple complementary observables

We introduce a novel correlation formula, termed the CMO (correlation of mass and observables) formula, which links nuclear masses to 12 distinct categories of experimental observables beyond mass data itself. These observables comprise four separation energies (${S}_{n}$, ${S}_{p}$, ${S}_{2n}$, ${S}_{2p}$), five reaction energies (${Q}_{{\ensuremath{\beta}}^{\ensuremath{-}}n}$, ${Q}_{\ensuremath{\epsilon}p}$, ${Q}_{d\ensuremath{\alpha}}$, ${Q}_{p\ensuremath{\alpha}}$, ${Q}_{n\ensuremath{\alpha}}$), and three decay energies (${Q}_{\ensuremath{\alpha}}$, ${Q}_{\ensuremath{\beta}}$, ${Q}_{2\ensuremath{\beta}}$). By integrating the CMO formula into an anisotropic kernel ridge regression (AKRR) algorithm, we predict the masses of 95 nuclei that are currently unmeasured across the nuclide chart. The predictive capability of this combined $\mathrm{AKRR}+\mathrm{CMO}$ framework was benchmarked against 121 nuclides whose masses are known in AME2020 but were listed as predicted values (marked $#$) in AME2003. This validation achieved a significant reduction in the root-mean-square deviation (RMSD) from 305 keV (using the WS4 model) to 106 keV. Our approach uses constraints from multiple complementary observables to improve prediction reliability. By moving beyond exclusive reliance on known masses, the $\mathrm{AKRR}+\mathrm{CMO}$ framework demonstrates improved extrapolation capability into structurally unexplored regions of the nuclear landscape.