Beyond Diamond: Interpretable Machine Learning Reveals Design Principles for Quantum Defect Host Materials

Solid-state spin defects in wide-bandgap semiconductors are leading candidates for quantum information processing, but systematic identification of suitable host materials remains limited by the cost of first-principles screening across vast chemical spaces. We address this with a composition-only machine learning framework built on heterogeneous Rashomon set ensembles: by contrasting the feature attributions of seven diverse classifiers, we extract consensus design rules that no single model identifies alone-filled valence s-, d-, and f-shells, low chemical heterogeneity, and enrichment in C, S, Si, and O favor quantum compatibility. Screening approximately 45,000 thermodynamically stable compounds, we identify 122 high-confidence candidates (confidence > 0.95), recovering most experimentally verified hosts (C, SiC, ZnO, ZnS) and predicting unexplored materials including TiO$_2$, PbWO$_4$, and layered chalcogenides (HfS$_2$, ZrS$_2$). Density functional perturbation theory calculations on 12 representative materials validate dielectric screening as a coherence proxy (R$^2$ = 0.89 against experimental T$_2$), and vacancy calculations for TiO$_2$ reveal deep, isolated mid-gap states favorable for spin-defect hosting. The framework provides transferable, physically grounded design principles for rational quantum materials discovery beyond traditional carbide and nitride hosts.