Um estudo comparativo das técnicas de predição na identificação de insucesso acadêmico dos estudantes durante cursos de programação introdutória

Abstract

The high failure rates of students in the introductory programming course within the universities worldwide have alarmed and worried many educators. Those rates can lead to losses of various types and interests. Thus, there are important reasons to try to clarify the main factors that possibly influence such failures. Furthermore, one of the major challenges is on how to early identify the students likely to in the introductory programming course, eventually allowing effective pedagogical interventions. Thus, in this study we aim to explore educational data mining techniques, in order to compare the effectiveness of prediction algorithms capable of identifying students likely to fail, in a timely manner suitable for pedagogical intervention. This study evaluated the efficacy of prediction algorithms in two different and independent data sources one in the classroom teaching mode and the other in the distance education mode in the disciplines in the introductory programming. The results showed that the techniques discussed in this study are effective in this task of prediction. In addition, it was shown also that after the completion of the pre-processing and adjustments to the parameters of the algorithms analyzed had an improvement in their results. At the end of the process, the Supported Vector Machine (SVM) algorithm showed the best results, both in the classroom teaching mode as in the distance, reaching an f-measure rate of 83% and 93% respectively.