Towards Multilevel Logistic Regression and Cluster Analysis for Academic Performance Examination and Prediction

Teaching approaches supported by Information and Communication Technologies, such as E-Learning, can provide valuable information from Learning Management Systems that generate real-time data about student interactions online. These data might be relevant to fit predictive models that can support educational institutions in making decisions and improving educational strategies. Based on this, several studies have proposed machine learning-based approaches to analyze students' academic performance, in particular, by using multilevel regression models that capture contextual effects and provide an analytical structure that supports the researcher in analyzing the model's behavior concerning the impact factors. Although the contributions on this topic have been quite significant, this article proposes an analytical approach that combines cluster analysis with a multilevel logistic regression model for the E-learning context. In this sense, the cluster analysis provided the formation of non-directly observable groups that proved to be significant in composing the multilevel structure of the logistic regression model. The study results support the researcher in analyzing the behavior of impact factors on academic performance considering the observed clusters. The analysis of the model's predictive capacity was also discussed, as well as considerations to direct future work.