Educational Data Mining for Dropout Prediction: an Experience at a University in Southern Brazil

Dropout is a problem that plagues public and private higher education institutions around the world and strategies for analyzing the reasons for the phenomenon abound in scientific publications. Many works that aim to find the most appropriate and effective techniques and practices for identifying dropout inducers in students end up being based on the use of technologies to improve data analysis and achieve a greater volume of processed information. The present study aims to identify good practices for the use of data mining for educational information. For this purpose, existing practices in the literature were investigated for structuring research with data from a public university in the interior of the state of Rio Grande do Sul. The study includes practical tests with the Decision Tree algorithms C4.5, Random Forest and Neural Networks in different datasets. The work demonstrates that the Random Forest algorithm was able to be more accurate in identifying students at risk of dropping out. From this experience other institutions will be able to base themselves for the definition of their best practices.