Decision trees for predicting the academic success of students

Abstract

The aim of this paper is to create a model that successfully classifies students into one of two categories, depending on their success at the end of their first academic year, and finding meaningful variables affecting their success. This model is based on information regarding student success in high school and their courses after completing their first year of study, as well as the rank of preferences assigned to the observed faculty, and attempts to classify students into one of the two categories in line with their academic success. Creating a model required collecting data on all undergraduate students enrolled into their second year at the Faculty of Economics, University of Osijek, as well as data on completion of the state exam. These two datasets were combined and used for the model. Several classification algorithms for constructing decision trees were compared and the statistical significance (t-test) of the results was analyzed. Finally, the algorithm that produced the highest accuracy was chosen as the most successful algorithm for modeling the academic success of students. The highest classification rate of 79% was produced using the REPTree decision tree algorithm, but the tree was not as successful in classifying both classes. Therefore, the average rate of classification was calculated for two models that gave the highest total rate of classification, where a higher percentage is achieved using the model relying on the algorithm J48. The most significant variables were total points in the state exam, points from high school and points in the Croatian language exam.