Prediction of student graduation grade with machine learning methods

For a quality education, it is a priority for higher education institutions to make correct and reliable decisions in the fields of management and education. As an example of the problems faced by higher education institutions in general, deficiencies or mistakes that may occur in academic planning, academically unsuccessful students, road maps for the future of students who will graduate can be shown. In terms of the quality of education, it is very important to solve such problems and take precautions. Thanks to the development of data mining and artificial intelligence methods, very high estimates can be made proportionally on the problems experienced, and as a result, meaningful solution-oriented results can be obtained. The fact that high-speed computers take more place in our lives, while the developed algorithms and Artificial Intelligence techniques are progressing rapidly, it is on the way to become a powerful tool for measures that can be taken academically, which will lead to important developments in the field of education, as in almost every sector. In the thesis study, the graduation grades of Bolu Abant İzzet Baysal University, Faculty of Economics and Administrative Sciences, Department of Public Administration students were estimated using machine learning methods Artificial Neural Networks, K-Nearest Neighbor Algorithm, Linear Regression, Support Vector Machines and Decision Trees. For this reason, the year-end grades of a total of 31 courses taken in the 1st and 2nd grades of 832 Public Administration Department students who registered and graduated between 2011-2018 were used. In the study, two different scenarios were created for the estimation of the graduation grade. In the first scenario, only the year-end grades of the first-year courses of the students and the graduation grade estimation were made, while in the second scenario, the year-end grades of the first and second-year courses of the students were used. In the study, it was seen that the model created with Artificial Neural Networks made more successful predictions than the others and the second scenario gave better prediction results than the first scenario.