8.4.3 Decision Support for Learning Engineering

The ultimate purpose of learning analytics is not to produce models but to improve decisions — decisions made by learning engineers during design, by instructors during delivery, by advisors during support, and by learners during their own study. Decision support for learning engineering is the applied science of ensuring that analytical outputs reach the right people in the right form at the right time to produce those better decisions.

Learning Analytics as Decision Infrastructure

The High-Leverage Opportunities report identified supporting human processes — including instructor decision support and predictive advising — as one of three strategic priority domains for the field.[LE-LS-GL-003] This reflects a hard-won insight from the field's first decade of deployment: technically sophisticated analytics systems frequently fail to change practice because the outputs are not designed for the decision-makers who need to act on them.

The design of decision support interfaces is therefore a core learning engineering competency, not a UX afterthought. A predictive model that outputs a risk score is incomplete as a decision support artifact: the learning engineer must also specify who receives the score, in what format, with what action-oriented recommendations, and with what escalation path if the recommended action is not taken. The Generalizable Learning Engineering Adoption Maturity Model articulates this as an organizational capability dimension — institutions at low maturity have analytics but no decision workflows; institutions at high maturity have closed loops from data to decision to outcome tracking.[LE-LS-AP-013]

Instructor Dashboards

Instructor dashboards are the most common form of LA decision support in institutional settings. Research presented at the LAK Conference and published in the Journal of Learning Analytics has examined what makes dashboards effective: the central finding is that dashboards that display only data without action affordances are far less effective than dashboards that directly link data insights to actionable interventions.[LE-LS-CE-002][LE-LS-JO-004]

Effective instructor dashboards typically:

Display student-level risk signals with clear thresholds that distinguish "monitor" from "act"
Identify the specific behaviors or knowledge gaps driving each student's risk level, not just the risk level itself
Provide direct communication tools (email, messaging) integrated into the dashboard workflow
Show which students have already been contacted and the outcome of prior outreach
Display class-level patterns (common error types, widespread skill gaps) that motivate course-level instructional adjustments

Ryan Baker's work at the Penn Center for Learning Analytics has documented that instructor adoption of dashboards is driven less by data volume or algorithmic sophistication and more by the perceived actionability of the information — instructors use dashboards when they believe the information will help them take actions that work.[LE-LS-PP-008][LE-LS-CO-004]

Recommender Systems for Content Sequencing

Recommender systems in learning analytics apply the methods of collaborative filtering and content-based filtering — developed for commercial recommendation engines — to the problem of selecting the next learning activity for each learner. Unlike commercial recommender systems optimized for engagement or purchase conversion, educational recommender systems must optimize for learning outcomes, which are harder to measure, have longer latency, and may require short-term difficulty or frustration in the service of long-term retention.

The Learning Engineering Toolkit describes recommender-based content sequencing as a component of adaptive learning systems that requires both a student model (to estimate current knowledge state) and a curriculum model (to map activities to learning objectives and estimate their expected value for learners in different states).[LE-LS-GL-007] The combination of these two models — a Bayesian network of knowledge components and a utility function over activities — constitutes the knowledge engineering core of an adaptive learning system.

Human-in-the-Loop Analytics

A persistent risk in automated decision support is the over-automation failure mode: systems that remove human judgment from consequential decisions about learners' educational trajectories. This risk is most acute in early-warning systems where automated flags can trigger escalating institutional responses without human review, and in adaptive systems where algorithmic content selection can systematically narrow the learning experiences of certain learner groups.

Human-in-the-loop (HITL) analytics design places human review at key decision points, using automation to handle routine cases efficiently while routing edge cases and high-stakes decisions to human experts. The U.S. Department of Education's AI in education policy report is explicit that automated decisions affecting individual learners' educational access, placement, or support must involve human review and must be subject to appeal.[LE-LS-GL-004]

Explainability Requirements

For decision support to be trusted and acted upon, the reasoning behind recommendations must be understandable to the people receiving them. An instructor who receives a flag that a student is "at risk" but cannot understand why the system flagged them cannot take targeted action. An advisor who receives a content recommendation but cannot explain to a student why this sequence was suggested cannot build the student's confidence and motivation.

Explainability requirements drive important technical choices in LA system design. Interpretable models (logistic regression, decision trees, BKT) have an inherent advantage over black-box models (deep neural networks) in decision-support contexts. Where black-box models are used for their predictive advantages, post-hoc explanation methods (SHAP values, LIME, attention visualization) can provide partial interpretability — identifying which features drove a particular prediction — though these explanations are themselves approximations that must be communicated with appropriate epistemic humility.

The LAK Conference has established algorithmic transparency and explainability as standing concerns for the field, with regular sessions devoted to evaluation of explanation methods in educational contexts.[LE-LS-CE-002]

Community Recommended Reading

Designing a Student Progress Panel for Formative Practice: A Learning Engineering Process — Rachel Van Campenhout, Michelle Selinger, Bill Jerome (2023, ICLS 2023)

ID: lecommons-rl-designing-a-student-progress-panel-for-formative-practice-a-learning-e · type: reading list item · confidence: medium · expert-validated: false

Why Did We Do That? A Systematic Approach to Tracking Decisions in the Design and Iteration of Learning Experiences — Lauren Totino, Aaron Kessler (2023)

ID: lecommons-rl-why-did-we-do-that-a-systematic-approach-to-tracking-decisions-in-the- · confidence: medium · expert-validated: false

Context enrichment applied 2026-04-17. See wrgr/lecommons for source data.

8.4.3 Decision Support for Learning Engineering

Learning Analytics as Decision Infrastructure

Instructor Dashboards

Recommender Systems for Content Sequencing

Human-in-the-Loop Analytics

Explainability Requirements

Further Reading

Policy, Reports & Grey Literature

Key People

Organizations, Conferences & Journals

Community Recommended Reading