Abstract
A data scientist could apply several machine learning approaches in order to discover valuable knowledge from the data. While applying several techniques, he might discover that some pieces of knowledge are invariant, what ever the technique he used. We consider such knowledge as mandatory concepts, i.e., unavoidable knowledge to be discovered. As interesting property, a mandatory concept is characterized by a non-shared isolated point, that relates pieces of data, e.g., an object to a property, a document to specific words, an image to a specific topic, etc. Hence, the isolated points allow to make the distinction between the concepts. In this paper, we present a new approach for mandatory concepts extraction by making a level-based properties composition. Hence, the N-Composites isolated points are identified and constitute a key element for mandatory concept localization. We experiment our new algorithm by considering the coverage quality metrics.