Towards Static-Analysis-Guided Grading of Student Code: Tool Design and Early Evaluation

Sabbir Hosen Mamun, Thorsten K. Wiegand

Workshop on Software Engineering Education and Training, pp. 12–18, Lisbon, Portugal · ACM

✦ Best Presentation

Conference2024

Abstract

Automated grading of student programming submissions typically relies on test-suite pass rates alone, missing structural and stylistic concerns that human graders would flag. We present a prototype that integrates static analysis metrics — cyclomatic complexity, coupling between objects, and naming-convention adherence — into a weighted rubric framework. A formative study with 48 student submissions shows the tool agrees with expert grader rankings at Spearman ρ = 0.81, with the largest disagreements occurring on submissions that pass all tests yet contain deeply nested control flow.

Cite this work

@inproceedings{Mamun2024Towards,
  title = {Towards Static-Analysis-Guided Grading of Student Code: Tool Design and Early Evaluation},
  author = {Sabbir Hosen Mamun and Thorsten K. Wiegand},
  year = {2024},
  booktitle = {Workshop on Software Engineering Education and Training},
  pages = {12–18},
  address = {Lisbon, Portugal},
  publisher = {ACM},
  doi = {10.1145/3643796.3648437},
  abstract = {Automated grading of student programming submissions typically relies on test-suite pass rates alone, missing structural and stylistic concerns that human graders would flag. We present a prototype that integrates static analysis metrics — cyclomatic complexity, coupling between objects, and naming-convention adherence — into a weighted rubric framework. A formative study with 48 student submissions shows the tool agrees with expert grader rankings at Spearman ρ = 0.81, with the largest disagreements occurring on submissions that pass all tests yet contain deeply nested control flow.},
}