This study investigates the application of machine learning techniques to predict students' final letter grades based on their midterm and quiz scores. The research utilizes a dataset comprising 5,001 students enrolled in courses taught by twelve faculty members. Following the application of predefined eligibility criteria, the final dataset consisted of 2,746 students. The AutoGluon framework, an Automated Machine Learning (AutoML) tool, was employed to train and optimize the models. The training process was conducted in two phases: first, hyperparameter tuning was performed on eleven machine learning models, and their performance metrics were evaluated. Subsequently, the four best-performing models were integrated into an ensemble model, which was retrained to enhance predictive accuracy. The ensemble model achieved a notable accuracy of 92.32%, demonstrating its effectiveness in predicting academic outcomes. This study underscores the potential of ensemble learning and AutoML in educational data mining, providing valuable insights for improving decision-making processes and supporting student success in academic settings.
Paşaoğlu et al. (Sun,) studied this question.