Description
Understanding chemical evolution in galaxies requires tracing how gas-phase metallicity evolves across cosmic time. Metallicity reflects a complex interplay between star formation, feedback, and gas flows. A key question is which galaxy properties most strongly influence metallicity. While stellar mass has long been considered the primary driver, recent studies suggest gravitational potential may be more important — yet no clear consensus has emerged.
To explore this, we perform a statistical analysis using machine learning on a large dataset combining ~5,000 galaxies from the MaNGA survey with measured dynamical masses and ~120,000 SDSS galaxies. We apply strict selection criteria to ensure high signal-to-noise emission line measurements, focusing on systems with reliable star formation diagnostics.
Using Random Forest regression and Partial Correlation Coefficients, we assess the importance of different galactic properties in determining metallicity. We also fine-tuned our machine learning model through hyperparameter optimisation to enhance the reliability of our findings. Interestingly, our results vary across the MaNGA and SDSS samples, revealing the sensitivity of outcomes to both the machine learning setup and choices made in deriving galaxy properties.
These findings highlight the challenges of isolating metallicity drivers, particularly regarding selection effects and methodological choices. While our analysis aligns with some prior results, it also reveals discrepancies, emphasising the importance of robust cross-comparisons. Ultimately, our study offers a local benchmark and a path forward: applying this framework to JWST observations will allow us to trace metallicity scaling relations across cosmic time and refine our understanding of galactic chemical enrichment.
