Inducing Fuzzy Models for Student Classification

We report an approach for implementing predictive fuzzy systems that manage capturing both the imprecision of the empirically induced classifications and the imprecision of the intuitive linguistic expressions via the extensive use of fuzzy sets. From end-users' point of view, the approach enables encapsulating the technical details of the underlying information system in terms of an intuitive linguistic interface. We describe a novel technical syntax of fuzzy descriptions and expressions, and outline the related systems of fuzzy linguistic queries and rules. To illustrate the method, we describe it in terms of a concrete educational user modelling application. We report experiments with two data sets, describing the records of the students attending to a university mathematics course in 2003 and 2004. In brief, we aim identifying the failing students of the year 2004, and develop a procedure for empirically inducing and assigning each student a fuzzy property "poor", which helps capturing the students needing extra assistance. In the educational context, the approach enables the construction of applications exploiting simple and intuitive student models, that to certain extent are self-evident.