Evaluating Instructor Effectiveness in Creative-Technology Education: A Machine Learning Approach with Longitudinal Data

Abstract

Instructor effectiveness in creative-technology education remains challenging to assess objectively due to the multidisciplinary nature of courses and reliance on subjective evaluation methods. This study presents a comprehensive analysis of instructor performance across eight course families using a novel longitudinal dataset of 755 student-course entries collected over five semesters (Fall 2023–Summer 2025) from the Department of Multimedia and Creative Technology at Daffodil International University, Bangladesh. All courses were taught by a single instructor, eliminating inter-instructor variability. This study employed four machine learning algorithms: Random Forest, Support Vector Machine, Logistic Regression, and K-Nearest Neighbors validated through 5-fold cross-validation to predict student grades and identify high-impact assessment components. Results demonstrate that Random Forest achieves the highest accuracy for laboratory courses (61.43%), while SVM performs best for theory courses (51.25%). Feature importance analysis reveals that final examinations dominate laboratory predictions (34.41%), whereas theory courses exhibit balanced importance across continuous assessments (26.48%), finals (24.79%), and midterms (20.37%). Statistical analysis using ANOVA confirms significant performance variation across course families (F = 3.82, p = 0.0004) ranging from 55.2% to 65.5%. This work establishes benchmarks for educational data mining in creative technology disciplines and provides empirical foundations for data-driven pedagogical interventions and early warning systems.