Data mining tool for academic data exploitation: selection of most suitable algorithms

Abstract

SPEET project is aimed at exploiting the potential synergy among the huge amount of academic data actually existing at universities and the maturity of data science in order to provide tools to extract information from students’ data. A rich picture can be extracted from this data if conveniently processed. The purpose of this project is to apply data mining algorithms to process this data in order to extract information about and to identify student profiles. In this document, the results obtained at SPEET project under the development of the data mining tools are presented. More specifically, two mechanisms have been developed: a clustering/classification scheme of students in terms of academic performance and a drop-out prediction system. The document starts by addressing the motivation of the development of data mining tools along with the considerations taken into account for academic data gathering. These considerations include the proposed unified dataset format and some details about confidentiality issues. Next, the students’ clustering and classification schemes are presented in detail. More specifically, a description of the considered machine learning algorithms can be found. Besides, a discussion of obtained results when considering data belonging to the different SPEET project’s partners is addressed. Results show how groups of clusters can be automatically identified and how new students can be classified into existing groups with a high accuracy. Finally, the implemented drop-out prediction system is considered by presenting several algorithms alternatives. In this case, the evaluation of the dropout mechanism is focused on one institution, showing a prediction accuracy around 91 %. Algorithms presented at this document are available at repositories or inline code format, as accordingly indicated.