This paper presents a graph-based grading system for Java introductory programming courses, eGrader. This system grades submission both dynamically and statically to ensure a complete and through grading job. While dynamic analysis is based on JUnit framework, the static analysis is based on the graph representation of the program and its quality which is measured by software metrics. The graph representation is based on the Control Dependence Graphs (CDG) and Method Call Dependencies (MCD). eGrader outperformed existing systems in two ways: the ability of grading submission with semantic-errors, effectively, and generating reports for students as a feedback on their performance and instructors on the overall performance of the class. eGrader is well received by instructors not only for saving time and effort but also for its high success rate represented by four performance measures which are sensitivity (97.37%), specificity (98.1%), precision (98.04%) and accuracy (97.07%).
[1]
Xiaohong Su,et al.
Semantic similarity-based grading of student programs
,
2007,
Inf. Softw. Technol..
[2]
John Waldron,et al.
Assessing the assessment of programming ability
,
2004
.
[3]
urs von Matt.
Kassandra: the automatic grading system
,
1994
.
[4]
Jack Hollingsworth,et al.
Automatic graders for programming classes
,
1960,
Commun. ACM.
[5]
Jéan H. Greyling,et al.
Marking student programs using graph similarity
,
2010,
Comput. Educ..
[6]
Paul Roe,et al.
Static Analysis of Students' Java Programs
,
2004,
ACE.
[7]
Christopher Douce,et al.
Automatic test-based assessment of programming: A review
,
2005,
JERC.
[8]
Paul Roe,et al.
ELP - A Web Environment for Learning to Program
,
2002
.