Predição de informações acadêmicas em dados de séries temporais

Abstract

Academic publications are one of the essential elements in the career of researchers. In fact, the quality and quantity of publications of researchers and research groups are a criterion to evaluate research projects by funding agencies, as well as the innovation potential of researchers, research groups and postgraduate courses. Given the importance of academic productivity for administrative decision-making, it is important to analyze the factors associated with the prediction of future publications. This work aims to investigate factors that may be related to the prediction of new quality publications, as well as strategies for predicting academic publications. Particularly, this work focuses on the treatment of information of researchers at Federal Institute of Goiás, considering factors such as academic production, researcher profile, previous publications, activities in teaching and in academic production groups. From such information, there was a temporal evolution of the production of each researcher. As a result, factors such as previous publications, activities related to research and subjects taught proved to be the most important to explain the evolution of research. The prediction of academic productivity using Random Forests is more effective than the previously employed Poisson Regression. Random Forests is also 10 times more efficient than Recurring Neural Networks (RNN and LSTM).