Machine Learning vs. SED-Fitting: A Comparative Analysis of Accuracy in Stellar Mass Estimation
Vahid Asadi, Akram Hasani Zonoozi, Hosein Haghi
TL;DR
This study tackles the persistent biases and computational bottlenecks in photometric stellar-mass estimation by performing a controlled comparison between a machine-learning approach (Pt-SNE) and a traditional SED-fitting method (LePhare). Using a Horizon-AGN COSMOS-like mock with known true masses, Pt-SNE is trained on noise-injected BC03 templates and evaluated against LePhare across 12 photometric bands, including analyses with band-reductions and varying signal-to-noise. Pt-SNE achieves substantially lower bias ($\text{Bias}=0.029$ dex) and a smaller RMS ($\sigma_F=0.169$ dex) than LePhare ($\text{Bias}=0.286$ dex, $\sigma_F=0.306$ dex), while also exhibiting superior robustness (lower $\sigma_{NMAD}$ and OLF) and requiring far fewer bands (12 vs 26) for comparable or better accuracy. Moreover, Pt-SNE is dramatically faster—approximately $3.2\times10^3$ times faster on large datasets—highlighting its suitability for scalable analyses in upcoming large-area surveys. The study also reveals limitations related to training-data representativeness, showing that extrapolation beyond the training domain can induce biases, thereby underscoring the need for diverse training sets and possibly broader template libraries for real-world applications.
Abstract
Traditional spectral energy distribution (SED)-fitting methods for stellar mass estimation face persistent challenges including systematic biases and computational constraints. We present a controlled comparison of machine learning (ML) and SED-fitting methods, assessing their accuracy, robustness, and computational efficiency. Using a sample of COSMOS-like galaxies from the Horizon-AGN simulation as a benchmark with known true masses, we evaluate the Parametric t-SNE (Pt-SNE) algorithm -- trained on noise-injected BC03 models -- against the established SED-fitting code LePhare. Our results demonstrate that Pt-SNE achieves superior accuracy, with a root-mean-square error (sigma_F) of 0.169 dex compared to LePhare's 0.306 dex. Crucially, Pt-SNE exhibits significantly lower bias (0.029 dex) compared to LePhare (0.286 dex). Pt-SNE also shows greater robustness across all stellar mass ranges, particularly for low-mass galaxies (10^9 to 10^10 solar masses), where it reduces errors by 47-53 %. Even when restricted to only six optical bands, Pt-SNE outperforms LePhare using all 26 available photometric bands, underscoring its superior informational efficiency. Computationally, Pt-SNE processes large datasets approximately 3.2 x 10^3 times faster than LePhare. These findings highlight the fundamental advantages of ML methods for stellar mass estimation, demonstrating their potential to deliver more accurate, stable, and scalable measurements for large-scale galaxy surveys.
