Filtering interlopers with photometry and diagnostic features: A machine learning framework validated with CSST slitless spectroscopy

TL;DR

This paper tackles interloper contamination in CSST and other slitless spectroscopic surveys by developing an XGBoost classifier that combines multi-band photometry with spectroscopic diagnostics to produce high-purity, high-completeness redshift catalogs. The model is trained on a 1e6‑source subset using 32 features (spectroscopic diagnostics, photometry, colors, and morphology) and optimized via a randomized 3‑fold CV search to achieve an AUC of 0.995 on validation, with a threshold of 0.7 balancing recall and contamination. On both test and full parent samples, it delivers a selection efficiency around 42% while retaining ~96.5–96.6% of retained galaxies as accurate redshifts and keeping the outlier rate at ~0.1–0.13%, demonstrating robust performance and stability. An interpretability SHAP analysis confirms physically meaningful feature contributions, particularly the importance of Hα and [O III] diagnostics and photometric colors, and ablation shows that photometry is crucial for maintaining low outlier and catastrophic interloper rates. The framework provides a practical path to high-fidelity large-scale redshift catalogs for CSST and similar surveys, with potential extension to Euclid and Roman in real data contexts.

Abstract

The slitless spectroscopic method employed by missions such as Euclid and the Chinese Space Station Survey Telescope (CSST) faces a fundamental challenge: spectroscopic redshifts derived from their data are susceptible to emission line misidentification due to the limited spectral resolution and signal-to-noise ratio. This effect systematically introduces interloper galaxies into the sample. Conventional strict selection not only struggles to secure high redshift purity but also drastically reduces completeness by discarding valuable data. To overcome this limitation, we develop an XGBoost classifier that leverages photometric properties and spectroscopic diagnostics to construct a high-purity redshift catalog while maximizing completeness. We validate this method on a simulated sample with spectra generated by the CSST emulator for slitless spectroscopy. Of the $\sim$62 million galaxies that obtain valid redshifts (parent sample), approximately 43% achieve accurate measurements, defined as $|Δz| \leq 0.002(1+z)$. From this parent sample, the XGBoost classifier selects galaxies with a selection efficiency of 42.3% on the test set and 42.2% when deployed on the entire parent sample. Crucially, among the retained galaxies, 96.6% (parent sample: 96.5%) achieve accurate measurements, while the outlier fraction ($|Δz|>0.01(1+z)$) is constrained to 0.13% (0.11%). We verified that simplified configurations which exclude either spectroscopic diagnostics (except the measured redshift) or photometric data yield significantly higher outlier fractions, increasing by factors of approximately 3.5 and 6.3 respectively, with the latter case also introducing notable catastrophic interloper contamination. This framework effectively resolves the purity-completeness trade-off, enabling robust large-scale cosmological studies with CSST and similar surveys.

Figures (6)

• Figure 1: Schematic workflow of the machine learning framework for constructing a high-purity, high-completeness galaxy sample. The diagram illustrates the pipeline from multi-source data preparation and feature engineering (combining photometric properties with spectroscopic diagnostics), through XGBoost classifier training and threshold optimization, to final evaluation and SHAP-based interpretability analysis. The output catalog is optimized for both high selection efficiency and high accuracy (high $f_{\mathrm{accurate}}$ and low $f_{\mathrm{outlier}}$ defined in section \ref{['sec:model_performance']}).
• Figure 2: Comparison of measured versus true redshifts for galaxies with valid redshift measurements. Left: Parent sample of galaxies with valid redshifts. Middle: Traditional SNR‑based selection: $\mathrm{H\alpha}_{\rm snr}\geq5\,{\rm OR}\, ([\mathrm{O\,{III}}]_{\rm snr}\geq5\,{\rm AND}\,\mathrm{H\beta}_{\rm snr}\geq 5)$. Right: Traditional combined‑criteria selection: the above SNR criteria plus ZWARNING = 0 and $\rm N_{lines}\geq3$. In each panel, the gray shaded regions denote the boundaries beyond which objects are classified as outliers ($|\Delta z| > 0.01(1+z_{\rm{true}})$). The title in each panel shows the total galaxy count and the selection efficiency (i.e., the fraction of selected objects relative to the total sample). The corresponding fractions of accurate measurements ($|\Delta z| \leq 0.002(1+z_{\rm{true}})$) and outliers are explicitly annotated.
• Figure 3: Receiver operating characteristic (ROC) curve of the model on the validation set. The area under the curve (AUC) is shown. The dashed diagonal line indicates random classification performance (TPR = FPR). The optimal threshold based on Youden's index (maximizing $\text{TPR}-\text{FPR}$) is shown as a solid orange circle. The blue and cyan hollow circles denote the default and a referenced threshold, respectively. The final selected threshold is indicated by a filled plus marker. The inset shows a magnified view of the region containing these threshold markers.
• Figure 4: Comparison of model performance on the test sample (left) and the parent sample (right). The upper subplot in each panel shows the measured redshift plotted against the true redshift for the corresponding sample. The lower subplots present the redshift offset, $\Delta z /(1+z_{\mathrm{true}})$, as a function of $z_{\mathrm{true}}$; this detailed offset analysis is only available for the test sample (left), due to the prohibitive size of the parent dataset. In both lower subplots, the mint‑blue solid line and coral dashed line represent the $\sigma_{\mathrm{NMAD}}$ scatter and the binned mean deviation, respectively.
• Figure 5: SHAP visualizations for the top 20 most influential features: (a) Feature importance ranking based on mean absolute SHAP values, quantifying average magnitude of feature contributions; (b) Beeswarm plot displaying the distribution of SHAP values for each feature, where dot color indicates feature value (red=high, blue=low), horizontal position shows impact direction (right=positive, left=negative), and density reveals value distribution patterns.