STRUCTURAL WHITE BOX AUTOGRADER USING CONTROL FLOW GRAPH SIMILARITY

• Main Author: Kamal Shafi, Muhammad
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/67162
• Institution: Institut Teknologi Bandung

Programming exercises are part of the programming learning process. To evaluate programming exercises, lecturers sometimes use an automatic grading system with black box testing as the basis. Testing with this technique can help evaluate the behavior of a program and can be implemented easily. However, there is a disadvantage in using black box testing, that is a solution can have only a little to no mistake at all in each tested scenario. A small difference in output format can give a logically correct program a small score. White box testing is an alternative that can be used as an automatic grading system which is more tolerant of errors. One aspect that can be tested using white box testing is the logical flow that can be represented as a control flow graph. Such assessment methods have been proposed before by several authors, but some still have shortcomings such as the accuracy that can still be improved, or that the system was still not ready to be used directly in our capstone system. In this thesis, a method is developed to overcome structural variations in programs as well as an assessment method based on the control flow graph structure by calculating the graph edit distance using an exact calculation approach and an approximation approach that also trying to take advantage of semantic information from program expressions. The result of the implementation is an automatic grading system and web service that can be used directly by the capstone system. The test result shows that the grading results of the assessment method proposed have an increased accuracy compared to previous method (Sendjaja et al. 2021) in predicting the manual grade of student exercise by looking at the correlation, mean absolute error, standard deviation of absolute error, average difference, and standard deviation of difference.