Avaliação de desempenho acadêmico utilizando redes neurais: uma análise exploratória de dados de rankings universitários

Abstract

The world lives in an era of knowledge and, due to a changing scenario, countries see in universities the possibility of being included in the world circuit of knowledge and skills. University evaluation has been the focus of several rankings worldwide and is an example of influence in the paradigm shift in universities, as the positive results from the evaluation process provide these institutions with a reputation, social and academic prestige, in addition to a benchmark among institutions about the services and practices provided in addition to the strengths and weaknesses of each course / university. In general, information on teaching, research and internationalization activities are combined to generate a grade used in a ranking order. This method is widely criticized for not showing unanimity in the formulation of its indicators and for relating the performance of an institution to just a number. In view of the above, this dissertation proposed an alternative means to academic ranking, the use of clustering techniques with neural networks. Self-organizing maps (SOM) are models of competitive neural networks. Through unsupervised learning, they perform a mapping between multidimensional data, generally two-dimensional, which approximates the original density of the information, being a technique widely used in areas such as data analysis and pattern recognition. This work presents a cross-sectional analysis of data from Brazilian universities through the training of maps with data from the 2014 and 2019 Ranking Universitário da Folha. From the profiles of the clusters, after the segmentation of the trained maps, it is possible to identify the positive points and of each group. With the identification of Higher Education Institutions (HEIs) in these different years, an analysis of the transitions between the clusters in the years 2014 and 2019 was carried out. Comparisons of the profiles of the clusters are shown in order to characterize their behavior in the analyzed period and showing a new one. As an alternative to the analysis of HEI performance data, the study also allows for the verification of disparities between the regions of Brazil.