Predicting Student Performance in Introduction to Information Technology Using Regression-Based Learning Analytics and Instructional Intervention Data

This study examines student performance across three instructional intervention strategies in an Introduction to Information Technology subject at a Malaysian private university in Perak. The dataset was obtained from the university system and involved a seven-week short-semester teaching period. A total of 308 diploma students were included and divided into three existing class groups: traditional lecture-based learning (n = 103), multimedia-assisted learning (n = 100), and blended learning (n = 105). Each group was exposed to a different instructional approach delivered by a different lecturer. Student performance was assessed through quizzes, assignments, midterm examination, and final examination. The study analyzed student performance using descriptive statistics, boxplot visualizations and formal statistical tests. The findings showed that the multimedia-assisted and blended learning groups generally performed better than the traditional lecture-based group in selected assessment components, particularly midterm examinations and assignments. However, no significant differences were found for quiz and final examination scores, indicating that performance differences varied according to assessment component. The correlation analysis showed that quiz and midterm scores had the strongest positive relationship among the assessment components (r = 0.474), while the assignment score had a very weak relationship with final examination score (r = 0.058). Blended learning showed weak positive correlations with all assessment components, while multimedia-assisted learning showed weak positive correlations with midterm and assignment scores. This is due to the limited sample size. The traditional lecture-based group showed negative correlations with midterm and assignment scores. In addition, this study applied regression-based learning analytics models to explore whether continuous assessment scores and instructional group information could predict students’ final examination performance. Eight regression techniques were tested and evaluated using mean squared error, mean absolute error, root mean squared error, and R² score. Based on the latest experimental results, KNN Regression produced the most favorable performance in terms of MSE (45.8526), RMSE (6.7715), and R² score (0.6467). These findings suggest that model performance depends on the data pattern, assessment structure, and prediction target. Therefore, the results should not be generalized to other courses or institutions without further validation. Semi-structured interviews with lecturers were also conducted to provide contextual insights into the implementation of the instructional interventions. Overall, the findings suggest that multimedia-assisted and blended learning approaches were associated with better performance in selected continuous assessment components. However, the findings should be interpreted cautiously since there are some limitations in collecting the dataset.