Data collection is one of the most critical and time-consuming tasks when dealing with Learning Analytics (LA) system design. This problem has been addressed in the context of METAL [6] (Modèles Et Traces au service de l’apprentissage de Langues), a project whose aim is to improve the learning experience of lower secondary education students of the Lorraine region in France. Our collection approach follows a hybrid methodology between a top-down Data Mining (DM) approach based on the CRISP-DM process [1] and a bottom-up analysis starting from the international e-learning specifications [9] Experience API (xAPI) [4] and OneRoster [5]. As a result of this hybrid methodology, we have developed METALRS [8], an extended Learning Record Store (LRS) that combines standard student learning traces with academic data and other more intrinsic student information in a multidimensional model.