Educational Data Mining Can Tell Us

EDM provides actionable insights that traditional "snapshot" assessments (like one-off exams) often miss.

EDM can identify "at-risk" students early by analyzing multidimensional predictors, including digital behavioral features and academic history (Aulakh et al., 2023; Chotchantarakun, 2025). Models using Random Forest algorithms have reached accuracy rates between 85% and 93% (Aulakh et al., 2023).

It reveals hidden trends in how different types of learners (e.g., field-dependent vs. field-independent) interact with software, allowing for "adaptive" learning experiences tailored to specific student needs (Angeli et al., 2017; Aulakh et al., 2023).

By analyzing which resources lead to success, institutions can optimize courseware, improve study schedules, and refine curriculum design (Aulakh et al., 2023; Chotchantarakun, 2025).

What Educational Data Mining Cannot Tell Us

Despite its power, EDM has significant "blind spots" regarding the human and theoretical side of education.

EDM often operates as a "black box," focusing on predicting behavior rather than providing a deep theoretical explanation of why a complex social phenomenon is occurring (Angeli et al., 2017). It identifies patterns (correlation) but does not inherently prove causation.

Data mining cannot replace "administrative intuition" or the qualitative understanding of an institutional context. Analytical tools are not "neutral reflections of reality" and require human interpretation to be effective (Tirado

Regulations like GDPR often require the "right to be forgotten." If students withdraw consent, longitudinal tracking is disrupted, meaning EDM cannot always provide a complete picture of a student’s long-term developmental trajectory (Papadogiannis et al., 2024).

Educational Data Mining (EDM) in Research: Insights and Limitations