Abstract
We consider functional outlier detection from a geometric perspective,specifically: for functional data sets drawn from a functional manifold whichis defined by the data's modes of variation in amplitude and phase. Based onthis manifold, we develop a conceptualization of functional outlier detectionthat is more widely applicable and realistic than previously proposed. Ourtheoretical and experimental analyses demonstrate several important advantagesof this perspective: It considerably improves theoretical understanding andallows to describe and analyse complex functional outlier scenariosconsistently and in full generality, by differentiating between structurallyanomalous outlier data that are off-manifold and distributionally outlying datathat are on-manifold but at its margins. This improves practical feasibility offunctional outlier detection: We show that simple manifold learning methods canbe used to reliably infer and visualize the geometric structure of functionaldata sets. We also show that standard outlier detection methods requiringtabular data inputs can be applied to functional data very successfully bysimply using their vector-valued representations learned from manifold learningmethods as input features. Our experiments on synthetic and real data setsdemonstrate that this approach leads to outlier detection performances at leaston par with existing functional data-specific methods in a large variety ofsettings, without the highly specialized, complex methodology and narrow domainof application these methods often entail.