Machine learning for the prioritization of Dyson sphere candidates

Dyson spheres are hypothetical megastructures proposed to partially or fully enclose a star and reprocess its radiation, producing mid-infrared excesses that may appear as rare anomalies in large stellar catalogues. No confirmed Dyson sphere is known, but the systematic prioritization of stars with unexplained photometric behaviour offers a practical way to support targeted astrophysical follow-up. In this work, we present a machine learning framework for the prioritization of potential Dyson sphere candidates. Rather than aiming at physical confirmation, the proposed approach ranks stars according to their similarity to a small set of literature candidates and to their deviation from well characterised normal stellar populations. The method is based on a symmetric two-phase learning scheme that integrates anomaly detection, metric-based few-shot learning and ensemble classifiers within a unified probabilistic ranking pipeline. Applied to Gaia DR3 and AllWISE photometry, the framework recovers previously reported mid-infrared excess sources and highlights new high-priority outliers that currently lack straightforward astrophysical interpretations. Among the tested models, isolation forests and prototypical networks with hard negative mining provide the most stable and effective rankings. These results indicate that data-driven prioritization can support large-scale technosignature searches by identifying anomalous stars that merit closer astrophysical investigation.