Predicting student achievement in BİLSEM exam with machine learning

This thesis examines the use and effects of machine learning algorithms in the process of identifying gifted students in Turkey. Using data from students who passed the first stage of the tablet application at the Science and Art Centers (BİLSEM) in Çankırı and Ankara, feature selection and model training were performed. This enabled the prediction of students who would be successful in the application. Among the 19 feature selection algorithms and 12 model training algorithms used, the combination of the Variance Threshold feature selection algorithm and the RadialSVM model development algorithm achieved a success rate of 0.67 in predicting successful students in the application. The findings indicate that these algorithms have high accuracy rates in predicting student success. Factors such as individual student achievements, the socio-economic status of families, parents' education levels, and the physical conditions of the school were found to have a significant impact on student success. Thus, a new decision support system that can be used in the BİLSEM selection process is proposed. The proposed system ensures a more objective and fair selection process by increasing accuracy in the analysis and modeling of student data. This way, it will be possible to identify gifted students at an early age and direct them to appropriate educational programs.