A Preprocessing Design Scheme for Sequential Pattern Analysis of a Student Database

In a data mining project evolved on a relational database often a significant effort needs to be done to construct the data set for the analysis. In fact, usually the database contains a series of normalized tables that need to be joined, aggregated and processed in an appropriate way to build the data set. This process generates various SQL queries that are written independently of each other, in a disordered manner. In this way, the database grows with tables and views which are not present at the conceptual level and this can yield problems for the development of the database. In this paper we consider a typical database containing data about students, courses and exams and illustrate some SQL transformations to build a data set to perform a sequential pattern analysis eventually combined with clustering and classification. In particular, we introduce in the student database some interesting patterns representing relationship between the exams given by students in various periods and the career of each student. This is achieved by introducing a particular encoding of a the career of a student. The resulting table can be analyzed with clustering and classification algorithms. We present a case study following this organization.