Mining Pre-Grade Academic and Demographic Data to Predict University Dropout

Abstract

Digital transformation is enabling institutions to enhance their processes by using data and technology. In education, digital transformation allows improving the learning experience as well as the institution processes. Within education 4.0, artificial intelligence applied to learning analytics is playing a key role for universities, particularly in the dropout issue, especially in STEM with the highest dropout rates. This is particularly relevant in the Latin American Higher Education scope, given the low labour productivity in these countries. In these countries, universities often have more demand than supply, and achieving an adequate balance between admission rates and dropout rates is a key issue. A high dropout rate harms the prestige of the university and damages students who were admitted without being adequate candidates. Understanding why students abandon their studies help to know what a university can do to avoid it. Data mining (DM) techniques can help discover the individual features that influence the dropout. There are different studies proposing models to predict dropout, and most are based on data that are not at the admission stage. We propose an approach that uses DM techniques to predict dropout based on data at the admission stage. We discover factors influencing dropout by a decision tree and association rules. We use a dataset of students of a computer science degree from a University in South America and achieve good performance when predicting dropout. The most attributes influencing dropout are the pre-grade performance in STEM subjects and the location of the city of residence.