Application of Artificial Intelligence using K-Means for Programming Question Assessment

Abstract

The manual assessment of programming assignments remains a significant challenge in educational settings due to its time-consuming nature and susceptibility to human error. Observational studies of course instructors reveal that over 40% have made grading mistakes, often due to fatigue or inconsistent evaluation standards. This study aims to develop an automated assessment system using artificial intelligence to enhance both objectivity and efficiency in the evaluation process. The method employed is the K-Means clustering algorithm, chosen for its ability to group answers based on similarities in logic and code structure rather than mere textual similarity. Five assessment categories were used as clustering standards: Logic and Algorithm, Data Structures, Object-Oriented Programming (OOP), Implementation, and Error Handling. The system was developed using an Agile Development approach and evaluated with student responses from programming courses. System performance was validated quantitatively by comparing cluster results against ground truth labels from manual grading. The system achieved 87% clustering accuracy, reduced the average grading time to 4.5 seconds per answer (compared to 13 seconds manually—representing a 65% efficiency gain), and decreased the inter-rater score standard deviation from 7.5 to 2.8 points. The results indicate that the system can deliver accurate real-time feedback. This study focused on programming questions ranging from easy to hard difficulty levels. In the future, the system could be enhanced by integrating advanced syntax analysis and expanding the evaluation criteria to support large-scale deployment.