Predição de desempenho em um ambiente virtual de aprendizagem utilizando algoritmos baseados em árvores de decisão e em regras.

Abstract

Data mining is the process that assists in decision making, performing the exploration of large amounts of data seeking to find anomalies, patterns and correlations. The use of data mining techniques allows the construction of predictive models to identify early a student with low academic performance. We used the Virtual Moodle Learning Environment of the Virtual UFPB, since it has real data of students. In order to predict academic performance, classification algorithms based on rules and decision trees were used. The decision of such choice is due to the relative ease of understanding and interpretation of decision rules and trees produced by these algorithms. Among the algorithms used for prediction, JRip obtained the best predictive performance, with accuracy of 88.9 %, the lowest accuracy was obtained by REPTree with a value of 80 %. Possibly useful rules were also obtained in predicting performance. The advantage of this approach is to inform teachers and educational tutors when a student is more likely to perform poorly, so preventive measures can be adopted.