Predicting Evasion Candidates in Higher Education Institutions

Since the nineties, Data Mining (DM) has shown to be a privileged partner in business by providing the organizations a rich set of tools to extract novel and useful knowledge from databases. In this paper, a DM application in the highly competitive market of educational services is presented. A model was built by combining a set of classifiers into a committee machine to predict the likelihood that a student who completed his/her second term will remain in the institution until graduation. The model was applied to undergraduate student records in a higher education institution in Brasilia, the capital of Brazil, and has shown to be predictive for evasion in a high accuracy. The unbiased selection of students with elevated evasion risk affords the institution the opportunity to devise mitigation strategies and preempt a decision by the student to evade.