Training dataset for machine-learning-based dynamical classification of near-Earth asteroids using short-duration backward integrations

This dataset contains time-series inputs used to train and evaluate machine-learning models for near-Earth asteroid (NEA) dynamical classification. The data are organized into class_0 (ejected) and class_1 (non-ejected). Each sample consists of a 2000-step time series of orbital elements (semi-major axis and eccentricity) representing 0.2 Myr of backward integration, with optional downsampling for model training. The time-series data are already in normalized form within the range [0, 1], obtained using min–max normalization. Labels are assigned using full 1 Myr integrations, indicating whether an object is dynamically ejected or remains bound. Additional details on preprocessing and usage are provided in the accompanying README and code repository.