Abstract
Student dropout is a complex problem that affects most post-secondary undergraduate programs, all over the world. The Industrial Engineering program of the ISVOUGA Institute, located in Sta. Maria da Feira, Portugal, is no exception. This research used a dataset containing students’ general information and the students’ marks for the already assessed courses. From this dataset, 17 potential predictors have been selected: five intrinsic predictors (gender, marital status, professional status, full/part time student, and age) and 12 extrinsic ones (the marks in all the 12 courses taught during the first two semesters of the program). The main goal of this research was to predict the likelihood of a student to dropout, based on the referred predictors. A binary logistic regression was used to classify students as having a high or low probability not to re enroll the program. To validate the appropriateness of the used methodology, the accuracy of the logistic model was compared, by means of a 5-fold cross-validation, to the accuracy of three classification methods commonly used in Data Mining: One R, K Nearest Neighbors, and Naive Bayes. Four variables were significant to the logistic model (the marks in Materials Science, Electricity, Calculus 1, and Chemistry). The two most influential predictors for student dropout are failing to pass in the less challenging courses of Materials Science and Electricity. Contrary to what we would think prior to this research, we found that failing in more challenging courses such as Physics or Statistics does not have a significant influence on student dropout.
Student dropout; Retention; Logistic regression; Data mining
Source: Authors.