Data-driven approximation of differential inclusions and application to detection of transportation modes

This article applies the Support Vector Data Description (SVDD) algorithm to approximate the graph of differential inclusions. It is proven that Gaussian SVDD can recover any compact graph if a large enough dataset is available. This data-driven approach can be used to identify discrete-valued parameters of nonlinear dynamical systems with unknown input signal. For illustration, the presented method is applied here both on real and synthetic data for detection of transportation modes based on linear velocity measurements.