We present a neural network-based method using Koopman Eigenfunctions to efficiently identify separatrices in high-dimensional dynamical systems, validated on synthetic, ecological, and neural network models.