Final-score-based learning evaluation often fails to comprehensively reveal students' achievement patterns and the quality of assessment instruments. This study applies an Educational Data Science approach to analyze examination data from the Computer Networks and Data Systems course, involving 142 students from four classes and 100 multiple-choice items. The analysis includes data preprocessing, exploratory data analysis, one-way ANOVA, Welch's t-test, item quality evaluation using Classical Test Theory (CTT), reliability testing, and topic mastery analysis. The results indicate no significant differences in student scores across classes (F = 1.20, p-value = 0.31) or instructors (t = 1.29, p-value = 0.20). Item analysis identified four difficult, 41 moderate, and 55 easy questions, with good instrument reliability (Cronbach's α = 0.80); however, 49% of the items exhibited poor discrimination indices. Items mapped to the Monitoring (Zabbix) topic showed the lowest average proportion of correct answers (41.7%). These findings suggest that combining statistical and psychometric analysis within an Educational Data Science framework can provide a more detailed picture to support learning evaluation and question bank improvement, although these results remain limited by the imbalance in student numbers across instructor groups, the uneven number of items across topics, and the heuristic, keyword-based topic classification.
Copyrights © 2026