A Binary Classifier for Identification of Female Evasion in Higher Level Technology Courses

Abstract

Courses related to the area of information and communication technology have a high dropout rate, roughly 40% in Brazil. It is essential to understand the underlying causes of this phenomenon to plan countermeasures to prevent it. This work proposes a binary classifier to detect patterns based on demographic and academic data of female students that indicate whether the student will complete the course. The most relevant attributes to determine evasion are identified. The anonymized data is obtained from the Bachelor in Telecommunications Engineering coordination at Universidade Federal Fluminense. Decision Tree (DT), K-Nearest Neighbors (KNN), and Support Vector Classifier (SVC) are used for building the classifier. The results show that the DT model performs better, on average, reaching an accuracy of 97%. The most important variables for identifying students at risk of dropping out are the number of completed semesters, academic performance coefficient, and obtained course load. To reduce classifier bias, the most influential variable is removed, and the results remain quite similar.