Prediction of the number of students who will take the make-up exam by extreme learning machine-based approaches

Make-up exams are an additional general exam that students who fail as a result of the general exams take, and these exams are performed by different universities in our country. The score obtained in the make-up exams is evaluated as the general exam score and is made after the general exams in order to evaluate whether the student has succeeded in the course. Depending on the grades obtained from the courses or the weighted grade averages, a student may not take the make-up exam. In cases where the students who will not take the exam are not determined in advance, an exam schedule is created, and question papers are issued with the assumption that all students who fail the make-up exam will take the make-up exam. As a result of this, both the waste of labor (sending the examiners to the exams) and the inefficient use of stationery materials such as paper and toner occur. In order to avoid these wastes, in this thesis study, non-personal data (distance, gender, general weighted grade point average, semester grade point average, midterm exam grade of the course, general exam grade etc.) belonging to each student were collected with a permission from the Konya Technical University Ethics Committee and the features that affect taking the exam were extracted. As a result, data sets were created using data from different courses and used in experimental studies. Whether or not each record in these datasets will take the make-up test has been considered as a classification problem and extreme learning machine-based approaches have been applied. In the experimental studies, firstly, basic extreme learning machine application was made in order to introduce the problem to the literature. The results obtained are also presented in this study, comparing the training and test accuracies of artificial neural networks trained with swarm intelligence algorithms and back propagation algorithm. The results showed that although the training set with the back propagation algorithm showed high success, both the swarm intelligence algorithms and the proposed approach were better in test sets. An experimental study includes the performance research of the hyper learning machine optimized with artificial bee colony on these datasets. In this experimental study, it is aimed to determine the number of hidden layer neurons and the activation functions used in more than one extreme learning machine, and the artificial bee colony algorithm is recommended to handle this process. When the proposed approaches and the results obtained are evaluated in general, determining the number of students to be integrated is effective in preventing waste and artificial neural network-based approaches offer acceptable performance for this classification problem.