Predicting student achievement with machine learning methods in transition from primary to high school exams in Bi̇leci̇k province

A good teaching process is directly proportional to the needs of students. In this sense, estimating students' academic success is also important for educational institutions. In recent years, as in many fields in our country, an increase in student achievements has been observed with the implementation of different projects in the field of education. Especially considering the increasing success in Turkish, Mathematics and Science courses, in order to maintain and improve the quality of education, important data and information belonging to students should be collected regularly and the results obtained by using multi-class machine learning methods should be evaluated by school administrations. In this study, a 32-question questionnaire with various demographic, social / emotional, and closed-ended questions, which are the variables expected to affect student performance, was applied to students studying in the 9th grade and 8th grade in 4 high schools that received students with an exam score in 2019. Achievement estimation has been made by combining student academic information based on paper pages and data obtained from the survey results and by creating a model with multi-class machine learning methods. First of all, by applying the feature extraction method, a subset of our initial attributes giving the best information was obtained. By comparing multi-class machine learning models created with these filtered attributes, it was observed that Random Forest method gave the best result.