Low-frequency noise control remains challenging because of the long wavelengths involved and the inherent limitations of conventional absorber designs. This study examines a Tesla valve–inspired acoustic metamaterial, referred to as the Asymmetric Coiled Channel (ACC), utilizing a computational and data-driven design framework aimed at enhancing low-frequency sound attenuation while maintaining suitability for cementitious, load-bearing systems. Thermoviscous acoustic simulations were conducted in COMSOL Multiphysics to account for viscous and thermal boundary layer effects within the ACC geometry. Key geometric parameters, including the coiling angle and channel topology, were systematically varied to investigate their influence on effective acoustic path length, resonance characteristics, and frequency-dependent sound absorption. The simulation results show that increasing geometric tortuosity shifts the resonance toward lower frequencies and enhances viscothermal energy dissipation, without requiring an increase in panel thickness. Among the configurations studied, the DET-90 design produced the lowest resonance frequency and the strongest low-frequency attenuation. Hybrid designs that incorporate Helmholtz resonator elements were also examined numerically and were found to provide targeted enhancement in the mid-to-high frequency range while preserving the low-frequency performance of the ACC. To efficiently explore the high-dimensional design space, a machine learning–based optimization framework was developed using the simulation dataset. Supervised learning models successfully captured the nonlinear relationships between geometric parameters and acoustic performance, enabling rapid prediction of sound absorption behavior across a wide frequency range. These models were further used to identify improved geometrical configurations beyond the initially simulated cases, substantially reducing the computational effort compared to exhaustive parametric studies. Overall, the proposed simulation-driven and ML-assisted framework offers a scalable approach for the computational modelling and optimization of acoustic metamaterials, with practical relevance for low-frequency noise mitigation in engineering applications, including railway infrastructure.