Inteligência artificial aplicada ao reconhecimento de padrões da evasão de estudantes da área de ciências exatas em uma instituição de ensino superior

Abstract

After a strong growth in the last two decades the market of the higher education goes through a period of stagnation, being necessary, in this context, the retention of the academics that already are part of its base of students. This work contributes in this scenario through the identification of the academics of the Area of Exact Sciences and Engineering of a Higher Education Institution of the Serra Gaúcha who are more likely to avoid studies using their database and Artificial Intelligence, with the Neural Networks model Artificial (RNAs). Firstly, financial, academic and social aspects were mapped from 2006 to 2018, followed by the data processing stage, where only the interruptions caused by curriculum change were discarded. The data were then processed in a perceptron topology RNA, in order to obtain a 93% accuracy in the identification of the tendency pattern to evasion of the studies (validation set). Also performed were by type of graduation, which have similarly satisfactory results in terms of pattern recognition. It is added to this evaluation the appropriate behavior of the Receiver Operating Characteristic (ROC) curve and the low correlation coefficients between factors and target matrix, which confirm that the RNA technique of backpropagation topology can be an important ally in the decisions of institutions regarding the management of student retention.