E-Evaluation of C programming assignments using generalized graph isomorphism and machine learning concept

Abstract

C program is a fundamental course in learning computer program. In both manual and e-learning platforms, assessment of students assignment homogeneously is a great challenge to the instructors. Several automated systems have been developed by using Static Analysis (SA), Dynamic Analysis (DA), Abstract Syntax Tree (AST), Graph Matching (GM), Output matching (OM) and many others. Evaluation accuracy is a very important concern to achieve student’s satisfaction and confidence. Existing systems evaluate a submission by exact structural matching, program’s output comparison and mapping with instructor’s solution. Therefore, performance of these systems is not so good, unable to evaluate submission program that is slightly modified from instructor’s solution and a submission having tiny error. To overcome these challenges, we have developed an automatic evaluation sys- tem by using TF-IDF, Generalized Graph Isomorphism and Levenshtein Token Edit Distance. We have also developed a solution corpus having maximum possible solutions for each problem, generated training program sets using solution programs to evaluate a new submission program automatically. The training program sets have been developed for three domains: prime number detection, bubble sort and binary search. The training programs also graded by three ex- pert human graders to ensure higher accuracy. The system has been tested by using real and synthetic datasets. We achieved 86% to 88% accuracy compared to expert human grader.