Prediction of the number of students taking make-up examinations using artificial neural networks

Abstract

Three different examinations for any course are primarily defined in higher education in Turkey: midterm, final and make-up exams. Whether a student has passed a course is decided by using the scores of midterm and final exams. If this student fails the course as a result of these exams, he can take a make-up exam of this course, and the score of the make-up exam is replaced with the final exam. However, some of the students do not take the make-up exam although it is expected that they take the make-up exam, due to different reasons such as average score, distance, low score of midterm exam, etc. Because the make-up exam plans and schedule have been performed in accordance with the number of students who failed the course, some resources such as the number of classrooms, invigilators, exam papers, toner are wasted. In order to reduce these wastages, we applied artificial neural networks, ANN, trained by different approaches for predicting the number of students taking make-up examinations in this study. In the proposed framework, some features of students and courses have been determined, the data has been collected and ANNs have been trained on these datasets. By using the trained ANNs, each student who fails the course is classified as positive (taking the make-up exam) or negative (not taking the make-up exam). In the experiments, the data of ten different courses are used for training ANNs by random weight network, error back propagation algorithm, some metaheuristic algorithms such as grey wolf optimizer, artificial bee colony, particle swarm optimization, ant colony optimization, etc. The performances of the trained ANNs have been compared with each other by considering training accuracy, testing accuracy, training time. BP achieves the best mean training accuracy on both unnormalized and normalized datasets with 99.36% and 99.7%, respectively. GWO achieves the best mean testing accuracy on both unnormalized and normalized datasets with 80.39% and 82.39%, respectively. Moreover, RWN has the best running time of less than a second for training the ANN on both normalized and unnormalized datasets. The experiments and comparisons show that an ANN-based classifier can be used for determining the number of students taking the make-up exam.