<![CDATA[The rapid development of Automated Machine Learning (AutoML) has transformed modeling practices in data science by automating preprocessing, feature selection, and hyperparameter tuning. However, its pedagogical implications in higher education remain underexplored. This study aims to compare the impact of AutoML and manual modeling approaches on students’ understanding of machine learning pipelines and model interpretability. A qualitative quasi-experimental design was employed involving final-year undergraduate students enrolled in a Data Science course. Participants were divided into two groups: one using AutoML tools and the other applying manual modeling procedures. Data were collected through in-depth interviews, learning observations, reflective reports, and artifact analysis of coding assignments. Thematic analysis was used to identify differences in conceptual understanding and learning experiences. The findings indicate that manual modeling fosters deeper structural comprehension of pipeline stages, including preprocessing, feature engineering, and evaluation mechanisms. In contrast, AutoML enhances efficiency and reduces technical barriers but tends to obscure internal modeling processes, potentially limiting interpretative insight. These results highlight important implications for curriculum design in data science education, suggesting the need for balanced integration between automation tools and foundational modeling practices.]]>
Copyrights © 2025
