Assessing student achievement with artificial neural networks: Analysis and recommendations

This study aims to evaluate the performance of an Artificial Neural Network (ANN) model trained to predict student achievement. The findings obtained as a result of the training and testing processes on a large data set are analyzed in detail in various important categories. It was observed that the model obtained low Root Mean Square Error (RMSE) values in courses such as Turkish and Mathematics, which are among the basic courses. This indicates that the model is able to make strong predictions in these subjects. However, an increase in test RMSE values was observed in subjects such as History of Turkish Revolution and Science, indicating that the model needs to be further improved in these subjects. The difference between training and test RMSE values in the prediction of mock exam scores is striking. This indicates that the model needs to be further improved in the prediction of mock exam scores. Furthermore, training durations were analyzed and different training durations were found for different categories. It was observed that the model should be trained for longer periods for more complex categories, especially for predicting mock exam scores. The main objective of this study is to emphasize that the ANN model exhibits a general effectiveness in predicting student achievement. In addition, the objectives of the study include; 1) Predicting High School Transition Exam (LGS) scores based on mock exams, 2) Measuring the similarity of trial exams to the LGS exam and 3) There are also goals to show that an academic advising/coaching infrastructure that determines the role of Artificial Intelligence (AI) in determining the academic success of students. These goals represent important contributions at the core of the study.