PREDICTING STUDENT PERFORMANCE IN A CORE ENGINEERING COURSE USING DECISION TREE METHOD

This study aims at using decision tree method to predict student performance in one of the core engineering courses: Strength of Materials. Three research questions are taken into consideration: 1) Can student performance be predicted by using Decision Tree? 2) Do a student’s score in prerequisite course, Semester and Cumulative GPA play a significant role in student performance throughout the related course? 3) Does Decision Tree predict more accurate than traditional regression techniques (Artificial Neural Network and Multivariate Linear Regression)? Total enrollment for the course was 93 students. A validated total of 372 data records were collected from 93 students, including 89 males and 4 females. Two of them are foreign students. Both gender and nationality, always of interest when examining data in core courses, was not included in the analysis since it is not eligible having only four female or two foreign students. Each student was associated with four data records. Data used in the analyses included the final grades of the core course Strength of Materials (ME271) and the prerequisite course Statics (ME251) and cumulative GPA in numerical values (0.0-4.0) and semester as discrete set of semester and year. These were obtained from the registrar’s office with ethics approval. Twenty-two of these were not included in the analyses because they either withdrew (14 students) or the researchers were unable to determine which grade they had taken from the prerequisite course (8 students), leaving a sample that consists of 71 students. Data belongs to only mechanical engineering students as the course is much more important for the mechanical engineering students than the students in other majors. The course selected for analysis was Strength of Materials (ME271) offered in the spring semesters of 2007, 2008 and 2009 at Marmara University - Faculty of Engineering. Student records were divided into the training dataset and the testing dataset. %70 of total data is selected as training dataset and the rest is selected as testing dataset. Applications and results of the methods have shown that decision tree is a powerful tool to predict the student performance in a core course. It is believed that this study will be helpful for researchers and lecturers as a gap is found in the literature.