Profiling and Archetyping of Higher Education Applicants Using Intelligent Data Analysis Techniques

Student dropout is a significant challenge in higher education, generating frustration in society and wasting resources. As a result, student retention constitutes a constant challenge for higher education institutions everywhere. This work focuses on the question: Can intelligent predictive data analysis techniques be applied to reduce the dropout rate in public and private universities? To answer this question, we have adopted an exploratory methodological approach based on historical data from approximately 13,715 applicants who later became university students. Unlike other research, based on publicly available data and statistics, our work relies on five years of actual data of students whose behavior has been synthesized in 27 variables related to socioeconomic, academic, and family factors and analyzes it. This paper has two main contributions. First, we propose intelligent predictive data analytics techniques and demonstrate that it is possible to profile and target the applicant for higher education as a strategy to reduce the dropout rate and improve their student welfare, so that the dropout probability can be used as part of an early warning in the recruitment process. Second, we propose a methodology for the segmentation and/or archetyping of applicants, which can be part of a corrective alert in the adaptation process. The profiling model and archetyping are replicable in private and public universities since we use easily extractable generic variables that do not require the university to have a high level of maturity in data management processes. Therefore, our results contribute to educational data mining (EDM), demonstrating that intelligent predictive data analysis techniques can be used to profile and archetype private and public university applicants for higher education. The evaluation of our solution proved that the neural network model profiled the dropout applicants with an accuracy higher than up to 97%, after which unsupervised learning was applied to generate archetypes.