Educational data mining is a process that aims at discovering patterns that provide insight into teaching and learning processes. This work uses Machine Learning techniques to create a student performance prediction model, using academic data and records from a Learning Management System, that correlates with success or failure in completing the course. Six algorithms were employed, with models trained at three different stages of their two-year course completion. We tested the models with records of 394 students from 3 courses. Random Forest provided the best results with 84.47% on the F1 score in our experiments, followed by Decision Tree obtaining similar results in the first subjects. We also employ clustering techniques and find different behavior groups with a strong correlation to performance. This work contributes to predicting students at risk of dropping out, offers insight into understanding student behavior, and provides a support mechanism for academic managers to take corrective and preventive actions on this problem.

Online streaming platforms have become the most important method of music consumption. Most streaming platforms provide tools to assess the popularity of a song by means of scores and rankings. In this paper, we present a methodology for predicting if a song will appear on Spotify's Top 50 Global ranking after a certain amount of time. Spotify is one of the biggest streaming services and if a song is popular on this platform, it is likely to ensure good financial return to the artists and their label. We approach the problem as a classification task and employ classifiers built on past information from the platform's Top 50 Global ranking, in addition to acoustic features from the songs. The Support Vector Machine classifier with RBF kernel reached the best results in our experiments with an AUC higher than 80% when predicting the popularity of songs two months in advance.

Concept drift, a phenomenon that can lead to degradation of classifier performance over time, is commonly addressed in the literature through detection and reaction strategies. However, these strategies often rely on complete classifier retraining without considering the properties of the drift, which can prove inadequate in many scenarios. Limited attention has been given to understanding the nature of drift and its characterization, which are crucial for designing effective reaction strategies. Drift descriptors provide a means to explain how new concepts replace existing ones, offering valuable insights into the nature of drift. In this context, this work examines the impact of four descriptors—severity, recurrence, frequency, and speed—on concept drift through extensive theoretical and experimental analysis. Experiments were conducted on five datasets with 32 descriptor variations, eight drift detectors, and a non-detection context, resulting in 1440 combinations. The findings reveal three key conclusions: (i) reaction strategies must be tailored to different types of drift; (ii) severity, recurrence, and frequency descriptors have the highest impact, whereas speed has minimal influence; and (iii) there is a need to incorporate mechanisms for describing concept drift into the strategies designed to address it.