Prediction of success with students' gain information in the programming course using machine learning and statistical methods

In the current education system, the success of the students is determined by exams and the deficiencies in the achievements are determined according to the scores they get from the exam and these deficiencies are tried to be compensated. However, it is not focused on how the identified deficiencies will affect the performance of the achievement targets to be measured in the next exams. Solution is sought according to the current situation, and plans are not made for the future to increase the student's performance. In this thesis, as the data set, in the 1st semester of the 2021-2022 academic year, The score distributions of 87 students studying in the 10th and 11th grades of Dr. Nureddin Erk-Perihan Erk Vocational and Technical Anatolian High School in the field of Information Technologies in the 3 exams applied in the Object Oriented Programming course were used. The questions asked in the exam were matched with the achievement titles in the course information form, and the performance rates of each student were tabulated according to the achievement titles. However, since the available data are scarce and in order to obtain more meaningful results, synthetic data has been produced by using the collected real data. The degree of closeness of the synthetic data to real data was confirmed by the detailed result report. Linear regression, k-nearest neighbor and decision tree algorithms supporting multi-output regression were used, since multiple achievement performances in the next exams would be estimated from the students' existing data. K-layer cross validation was applied to evaluate the success of the algorithms used. MAE, MSE, R2 and standard deviation were used for performance measurements. For the solution of the overfitting problem, the performance was improved by finding the best parameter values in the KNN and decision tree algorithms. According to the results, the best performance values were obtained with the KNN algorithm. Better results can be obtained by increasing the number of actual data and approximating the score distributions in the achievements measured in the exams. As a future study, an online system can be designed to prevent the student's failures in the next exams. This system, which is developed according to the performance estimates produced, can provide guidance and advice to the student. By using gamification, it can be ensured that the student learns while having fun by attracting attention.