A new classification model based on machine learning algorithms to predict student success

Today, the use of data-based machine learning algorithms, which are used to reveal important information from large data piles, is increasing. One of the most important areas in which machine learning algorithms are used is education. Countries with the understanding of modern education system determine the deficiencies in the process while making plans for the future based on the data obtained in the field of education. In this study, first of all, a research was conducted on the variables that affect students' exam success. After determining the variables affecting student success, an item pool was created based on these variables. This item pool was finalized after expert opinion and preliminary practices, and was reduced to 84 items. Students enrolled in schools in the central districts of Antalya that accept students with the High School Entrance Examination (LGS) score are considered successful, while students who attend schools with only a grade point average are considered unsuccessful. Data were collected from 613 students, 363 unsuccessful and 250 successful students, through an online survey due to COVID-19. The analysis part was started with the collected data. First of all, the items with a high degree of importance were determined, and the research continued on the answers given to the 30 items with the highest degree of importance. Items such as "the number of books belonging to the student", "the socio-economic status of the student", "the participation of the family in school activities" have a high degree of importance in student success. Since the data obtained within the scope of the study are categorical data, when the studies in the literature are examined, C5.0, CART, SVM and Random Forests algorithms, which are seen to work better with categorical data, were used in the analysis of the data. The data set is divided into two as training (80%) and test (20%) sets. Cross validation was applied for all algorithms to increase the error-free rate of the analysis. According to the model evaluation criteria obtained, the Random Forests Algorithm was found to be the most successful estimation algorithm. The estimations of the model coefficients for the classifier model were determined through the random forests algorithm. Through this new classifier model, the classification of success will be made for the student who will be involved in the process later. It is thought that the use of this new classifier model obtained is not only whether the students preparing for the LGS exam will be placed in a qualified school, but also that the deficiencies of the students can be detected early in the education period, these deficiencies can be eliminated and a positive contribution will be made in terms of success.