| Popis: |
This paper analyzes the contributions of features widely used in the automatic classification of students’ academic performance. In this classification problem, the relationship between various features and classifiers is analyzed using an exhaustive feature selection strategy. In this way, the optimal subset of features providing the highest classification performance is obtained. For this purpose, an academic performance dataset consisting of 15 distinct features and 480 samples is used. The features mainly belong to four different categories, including demographic, academic background, parent participation, and behavioral. The samples are from three different classes corresponding to the low, middle, and high levels of students’ success. For evaluations, 10 different classification algorithms are employed. Extensive experimental analysis reveals that the accuracy of the classification of students’ academic performance can be improved up to 79.40% using only 8 features rather than all. |
