Baker and Yacef (2009) conducted a significant early review of EDM. They summarized how researchers use log files, clickstream data, and interaction patterns from digital learning environments to study student learning behaviors. The study covers common EDM techniques, including classification, clustering, association rule mining, and sequential pattern mining.

What the Research Shows

Using EDM, the researchers uncovered patterns such as:

- Student engagement levels (e.g., identifying when students are gaming the system or off-task)

- Predictors of success or failure based on early activity patterns

- Knowledge mastery using models like Bayesian Knowledge Tracing

- Common error patterns made during problem-solving

- Sequences of actions that lead to productive learning

These findings come from analyzing detailed interaction data, including every click, hint request, wrong attempt, or time spent on tasks.

What Educational Data Mining Can Tell Us

1. Learning Behaviors and Patterns

EDM can detect behaviors such as guessing, procrastination, gaming the system, and persistent misconceptions (Baker, 2007).

2. Predictive Indicators of Performance

It can forecast which students are likely to pass, fail, or drop out based on early activity data (Romero & Ventura, 2020).

3. Personalized Learning Needs

EDM can identify the skills students have mastered or struggle with. This helps create adaptive learning systems and targeted interventions.

4. Process-Level Insights

Unlike traditional assessments, EDM captures how students learn, not just whether they get the right answer. This includes strategy use, response times, and learning trajectories.

What Educational Data Mining Cannot Tell Us

1. Internal Psychological States

EDM can identify off-task behavior but cannot fully explain why a student is disengaged. The reasons could include boredom, anxiety, confusion, or external distractions (Winne, 2017).

2. Complex Human Qualities

Skills like creativity, resilience, empathy, and collaboration are hard to measure using log data alone.

3. Context Behind the Numbers

EDM shows patterns, but understanding the context requires human interpretation. For example, low activity might indicate poor internet access rather than low motivation.

4. Long-Term Learning or Transfer

Most EDM studies focus on immediate, platform-based behaviors. They cannot reliably measure long-term retention or the ability to apply skills in real-world situations.

5. Equity or Cultural Factors

EDM systems may reproduce biases if their training data reflect existing inequalities. EDM cannot detect socio-cultural influences on learning outcomes.

Baker, R. S., & Yacef, K. (2009). The state of educational data mining in 2009: A review and future visions. Journal of Educational Data Mining, 1(1), 3–17.