Abstract: Interest for applying machine learning in astronomical object classification has been growing with forthcoming huge surveys. Here, we put forward a methodical approach to analyzing variable star light curves through the application of Koopman theory-based modern data-driven techniques for dynamical system analysis. We employ this method on light curves associated to RRLyrae stars in the Galactic globular cluster $\omega$ Centauri. Curves are thus summarized by a handful of complex eigenvalues, corresponding to oscillatory or fading modes. In contrast with RRab variables, we find that RRc variables are defined in terms of fewer eigenvalues, which reflects the simpler structure of their light curves. Additionally, we show how Blazhko variables may be identified using DMD eigenvalues and that a physical interpretation of the related modes may be obtained in terms of the Blazhko effect.