Constraining Axion-Like Particle mediated Dark Matter with Observational Constraints: A Statistical and Machine Learning Approach

Abstract: We present a comprehensive investigation into the phenomenological consequences of axion-like particle (ALP) mediated dark matter (DM) on neutron star (NS) structure. Using a relativistic mean-field framework with non-linear mesonic self-interactions constrained by nuclear data and astrophysical observations, we explore the DM parameter space spanning $m_\chi \in [0, 1000]~\mathrm{GeV}$ and $q_f \in [0, 0.06]~\mathrm{GeV}$, generating over 30,000 equations of state (EoSs). Two representative hadronic EoSs are employed, a stiff (EoS1) and a soft (EoS18), with explicit inclusion of the crustal EoS. A multi-tiered statistical filtering scheme, combining voting, likelihood, and kernel density estimation scores, is applied using constraints from radio and X-ray pulsars, GW170817, and the low-mass compact object HESS J1731-347. We find that models satisfying the PSR J0614$-$3329 radius bound automatically comply with HESS, positioning ALP-mediated DM as a viable explanation for low-mass compact objects while still supporting $2\,M_\odot$ NSs. For the stiff EoS, we obtain $m_\chi \gtrsim 43~\mathrm{GeV}$, with score-weighted posteriors favoring $q_f = 0.034^{+0.020}_{-0.012}$ and a broad allowed DM mass range $m_\chi \in [101, 949]~\mathrm{GeV}$ (median $\sim 466$ GeV). The soft EoS yields no strict lower bound, though large $m_\chi$--$q_f$ combinations are disfavored. A high-precision supervised regression model built with AutoGluon achieves $R^2 > 0.998$ for inferring DM parameters from NS observables. Feature analysis reveals $m_\chi$ is constrained by structural ratios such as $R_{1.6}/R_{1.4}$, whereas $q_f$ is set mainly by the tidal deformability $\Lambda_{1.4}$.