LSST Simulated Supernova Classification Dataset (Ia vs Core-Collapse)

Simulated supernova survey data from the Vera C. Rubin Observatory (LSST) pipeline, packaged for machine learning classification tasks. Task: classify Type Ia supernovae (SNe Ia) vs core-collapse supernovae (CC SNe) from tabular light-curve fit parameters and host galaxy properties. Training set (~2.37M rows): biascor simulations from the DESC LSST ELASTiCC pipeline (Ia + CC combined, shuffled). Test set (~5,559 rows): events that passed the full survey pipeline including SALT2 light-curve fitting and BBC bias corrections. The PROB_SCONE_PREDICT column provides the output of the deployed SCONE neural-net classifier (baseline accuracy 84.4%) for direct comparison. Features (30 columns): SALT2 fit parameters (x1, c, mB), fit quality (chi2, FITPROB), per-band signal-to-noise, host galaxy magnitudes in 6 LSST bands (ugrizy), host stellar mass, SFR, colour, and photometric redshift. Simulations produced with: SNANA (Kessler et al. 2009, PASP 121, 1028). Pipeline: DESC LSST Analysis-2 (Pippin framework). Classifier baseline: SCONE (Qu et al. 2021, AJ 162, 67). See README.md for full column descriptions, suggested projects, and quick-start Python code.