A comparative study of dynamic mode decomposition (dmd) and dynamical component analysis (dyca)

Abstract

Two dimensionality reduction methods, dynamic mode decomposition (DMD) and dynamical component analysis (DyCA), are briefly introduced and compared by application on epileptic EEG data. DMD approximates a linear operator whose eigendecomposition describes the underlying system in frequency space. A reduction in dimensionality is achieved by retrospectively selecting relevant DMD modes. DyCA, on the other hand, naturally provides a dimensionality reduction by projecting onto a relevant subspace in time domain during the process.