Predicting success by examining the video watching behaviors in flipped classrooms using data mining

In this thesis, it was aimed to model students' success levels via the classification methods, which are included in data mining methods, by using the data of video watching behaviors in the flipped classrooms. Prediction of success is important to design the learning environment more efficiently and to give individual feedback. Estimates made by using large amounts of data from online environments have become very popular in educational data mining. In the present study, video watching data were obtained from 404 university students in different departments enrolled in Basic Computer Applications course for 4 semesters. Before the face-to-face education, the researcher prepared 11 videos compatible with course content with the average of 15 minutes. During the 14-week course period, students watched these videos through the Moodle learning management system. To record students' data related to number of jumping, number of logins, number of different video views etc., a tool built on the learning management system was used. 11 qualities collected from digital environment and 6 demographic information collected via the survey method, totally, 17 qualities were obtained in data set. Success performance of the students were determined as successful and unsuccessful. Random Forest, Support Vector Machines and Naive Bayes algorithms, which are among the most preferred classification algorithms in success prediction studies in the field to reveal hidden patterns in the data, were used in the analysis phase to answer the main problem of the study. Since the data set was unstable, synthetic data were created via the smote technique, and cross validation method was also used. When the obtained results were examined, it was revealed that the course success could be predicted with a high accuracy rate. The algorithm with the most successful classification rate was the Random Forest algorithm with a total accuracy rate of 83.11%. Support Vector Machines were able to predict with a very low accuracy rate compared to Naive Bayes and Random Forest algorithms. Considering the fact that success levels of unsuccessful students can be increased before final exams, prediction of unsuccessful students is thought to be more important than prediction of successful students. It was determined that the Random Forest algorithm with a accuracy rate of 87,67% in predicting unsuccessful students achieved a much higher correct prediction than other algorithms used.