RCA as a Data Transforming Method: A Comparison with Propositionalisation

This paper aims at comparing transformation-based approaches built to deal with relational data, and in particular two approaches which have emerged in two different communities: Relational Concept Analysis (RCA), based on an iterative use of the classical Formal Concept Analysis (FCA) approach, and Propositionalisation coming from the Inductive Logic Programming community. Both approaches work by transforming a complex problem into a simpler one, namely transforming a database consisting of several tables into a single table. For this purpose, a main table is chosen and new attributes capturing the information from the other tables are built and added to this table. We show the similarities between those transformations for what concerns the principles underlying them, the semantics of the built attributes and the result of a classification performed by FCA on the enriched table. This is illustrated on a simple dataset and we also present a synthetic comparison based on a larger dataset from the hydrological domain.