Abstract Type Ia supernovae (SNe Ia) are key tools for addressing major cosmological questions like the Hubble tension and dark energy. Modern surveys are predominantly photometry based, making photometric classification of SNe Ia essential for precision cosmology. We evaluate whether functional principal component analysis (FPCA) scores derived from light curves, combined with ensemble learning, can reliably distinguish SNe Ia from other transients using a subset of PLAsTiCC sources with well-sampled light curves in a fully photometric approach. FPCA offers a data-driven, flexible characterization without strict model-based assumptions. Light curves are fitted by minimizing residuals with penalty terms from clean samples, ensuring robustness to outliers and poor band sampling. The first two FPCA scores and peak magnitudes in five Legacy Survey of Space and Time (LSST) bands serve as classification features. We implement two binary classifiers: an ensemble boosting model (CatBoost) and a statistical probabilistic method based on Euclidean distances. CatBoost slightly outperforms the statistical method, achieving 96.5% accuracy and 96.6% precision. Both accuracy and precision remain satisfactory (>90%) for photometric redshift uncertainties up to σ = 0.05. On the spectroscopic DES Y5 sample, both methods maintain ∼90% accuracy and 95% precision, demonstrating excellent cross-survey generalization. Applied to DECam DDF and DESIRT transients, predictions strongly agree, and their intersection yields a high-confidence SN Ia sample for cosmological analyses. This FPCA-based framework provides a powerful, flexible tool for transient classification in upcoming surveys such as LSST and Roman.
Reza et al. (Thu,) studied this question.