Abstract:

Most times, students’ performance generally might fluctuate across semesters or years. Sequel to this, it is of utmost importance to pin point those various factors that could cause this fluctuation as well as predict students’ performance by tuning these factors. In this research work, a proper classification work flow was followed to predict students’ academic performance using the student performance dataset from the UCI data repository. This dataset was preprocessed accordingly, and proper data visualization techniques were applied in other to get a better understanding of the dataset. In addition, a form of cross validation which entails applying more than one classification models (i.e., Logistic regression using deep learning, Naïve Bayes and Recurrent Neural Network (RNN) using the Gated Recurrent Unit (GRU)) were performed in order to conclude on the best classification model based on accuracy produced by each model. The result of this research shows that the Recurrent Neural Network model produced the best accuracies for both the Mathematics and the combination of Mathematics and Portuguese datasets. It seems so since the Gated Recurrent Unit cell allows the recurrent network to save more historical information for a better prediction. On the other hand, the Logistic regression model produced the best accuracy for the Portuguese dataset, since it performs best when the data is distributed in a fashion such that it can be linearly classified.

Keywords:

Students' performance, prediction, supervised learning.