Understanding the link between ecological and morphological features in extant species is a key issue, notably in paleontology since it allows the inference of extinct species ecologies from their morphological features. Such predictions are classically done using linear discriminant analysis (LDA): this method fits a model on a training set containing individuals for which both categorical traits (e.g. ecological classes) and continuous traits (e.g. morphology) are known, to be then able to predict the class of an individual for which only continuous traits are known. Morphology is not the only signal which can help to infer past ecologies: the phylogenetic position of an individual can also be highly informative since closely related species often share the same ecology. Moreover, with the rise of 2D and 3D geometric morphometrics, datasets with more traits than species (high-dimensional datasets) are now commonplace, but classical discriminant analysis methods significantly lose statistical power when the number of morphological traits (p) approaches the number of species (n), and are not even computable when p is higher than n. Here we develop a new discriminant analysis which is both phylogeny-informed and applicable to high-dimensional datasets through penalized likelihood techniques. The performances of this newly implemented method were assessed on simulated and empirical datasets. It appears that this new method outperforms, in many situations, conventional discriminant approaches when applied to comparative datasets (e.g., phylogenetically related species).