A Bayesian Network Model for Predicting Exoplanet Habitability Probabilities

The classical habitable-zone (HZ) criterion screens planets by radiative equilibrium alone and omits processes that decide whether a surface stays habitable over Gyr timescales: atmospheric retention (set by mass), climate stability (set by orbital eccentricity), atmospheric loss (driven by stellar activity) and temperature class (set by insolation flux S). Insolation separates Earth from Venus, which have comparable mass, nearly circular orbits and the same host star, yet Venus lies inside the runaway-greenhouse inner edge. This paper constructs a four-factor Habitability Potential Index (HPI) and a discrete Bayesian network whose nodes are planet mass M, eccentricity e, a stellar-activity index A and insolation S. Nonlinear maps convert each observable into an interpretable sub-factor, and a weighted sum gives a reproducible deterministic rank. The network infers P(habitable | data) exactly by enumeration, and missing observables are marginalised rather than discarded. Principal-component analysis shows that the four observables are essentially uncorrelated, with variance split almost evenly (26.2, 25.1, 24.6 and 24.0 per cent), which justifies using all four as network nodes. A random-forest regression against HPI recovers the expected importance order, with mass and insolation leading. Applied to an archive-like sample of 2795 planets, the framework yields a sparse high-habitability tail and a conservative posterior that compresses risk more sharply than the linear index. A Solar System check assigns Earth P ≈ 0.61 and Venus P ≈ 0.15. The combination of a physics prior and exact probabilistic inference provides an extensible ranking layer for target selection in large exoplanet catalogues.