Abstract Unambiguous identification of Majorana zero modes in topological superconductors remains a challenge due to complex in-gap states that can also produce zero-bias conductance peaks. Here we demonstrate a data-driven workflow that integrates pixel-wise spectral deconvolution with machine learning to analyze tunneling spectroscopy from an intrinsic topological superconductor. Local density of states spectra, acquired with a millikelvin scanning tunneling microscope under magnetic fields, are decomposed into multiple Lorentzian peaks. The extracted peak parameters are assembled into a structured feature set and clustered without supervision. The clustering separates vortices in the superconductor exhibiting zero-bias-peaks consistent with established characteristics of Majorana zero modes from vortices displaying zero-bias-peak-mimicking features of trivial origin. Spatially resolved zero-bias-peak distributions differentiate isotropic vortex cores with well-defined peaks from vortices that exhibit locally distorted peaks. Comparing these distributions to defect locations measured without magnetic field, we find a correlation between local heterogeneity and their formation. This objective and reproducible workflow advances reliable detection of Majorana zero modes, providing a foundation for their manipulation towards quantum computation.