Application of Data Mining Methods for Classification of Student Results from Conveying an ECTS Faculty Course

Abstract

In the large datasets (tables) containing grades gained by students according to many criteria (attributes, i.e., columns in the tables) for particular ECTS courses, it is a common situation every criterion to have its own sub criteria, and these sub criteria to be linearly combined to form the corresponding criterion. Therefore, practically, for each student we get a single record (single row in a table) with many attributes. The main goal of our research was to assess how well selected data mining methods are capable of detecting the linear dependency of the final course grade from the course criteria. To this purpose, we collected a dataset containing results from a particular course held at our institution, and we made appropriate experiments. We evaluated three different data mining methods on this dataset (in its raw form) in order to discover how well they would be able to model the criterion for forming the final grade, and to estimate the classification accuracy that they would achieve on independent test sets. In this paper we study the performance of these data mining methods on the dataset, analyze the results and point out further directions for research.