8.4 Artificial Intelligence and Machine Learning in Learning Analytics

Artificial intelligence and machine learning techniques have progressively expanded what learning analytics can detect, predict, and optimize. Section 8.4 surveys this landscape: the distinct AI paradigms in use, the machine learning algorithms that have proven most productive in educational data contexts, and the decision-support frameworks that translate AI outputs into actionable guidance for learning engineers, instructors, and institutional leaders.

AI in Learning Analytics: A Layered History

The application of AI to educational data has a layered history that mirrors the broader arc of AI development. The first generation — rule-based AI and production-rule systems — emerged from cognitive science and informed the first intelligent tutoring systems of the 1980s and 1990s.[LE-LS-AP-001] The second generation — probabilistic and Bayesian methods, exemplified by BKT — dominated the 1990s and 2000s and remains the most widely deployed student-modeling approach in production systems.[LE-LS-AP-011] The third generation — deep learning, reinforcement learning, and large language models — has produced rapid methodological advances since roughly 2015 and is reshaping both the technical possibilities and the ethical obligations of learning analytics practice.

Each generation extends rather than replaces its predecessors. Production-rule models still provide the cognitive scaffolding for expert knowledge representation in adaptive systems.[LE-LS-AP-014] Bayesian methods still provide the interpretable, data-efficient student models that most institutional deployments depend on. Deep learning methods extend these capabilities to larger datasets, richer behavioral signals, and natural language inputs. Learning engineers must understand all three generations to make informed architectural decisions.

The Role of the Simon Initiative and Open Infrastructure

The Simon Initiative at CMU has been a central force in connecting AI and ML advances to learning engineering practice — not just through research but through open infrastructure that enables the field to study and replicate findings at scale.[LE-LS-CO-002] DataShop, the Open Learning Initiative, and the LearnSphere analytics workbench together provide the data, tools, and platforms that allow learning engineers to develop, test, and deploy AI-based analytics without building infrastructure from scratch.

This infrastructure matters for AI/ML adoption because educational datasets are typically smaller, noisier, and more fragmented than the datasets on which AI advances are demonstrated in commercial contexts. The Simon Initiative's investment in shared data infrastructure and open tools lowers the barrier for applying ML methods responsibly and reproducibly.

The Learning Engineering Toolkit's AI Framing

The Learning Engineering Toolkit positions AI and ML as tools that serve the iterative LE process, not as ends in themselves.[LE-LS-GL-007] This framing has two practical implications. First, the choice of AI method should be driven by the learning question being asked — which requires understanding what different AI paradigms can and cannot answer. Second, AI outputs must be evaluated not just on technical accuracy metrics but on whether they improve learning outcomes when deployed — which requires empirical study in the learning context itself.

The U.S. Department of Education's AI in education policy report reinforces this point from a governance perspective: AI systems deployed in educational contexts must be held to higher standards of transparency, equity, and accountability than AI systems deployed in commercial contexts, because the stakes for learners — particularly learners from historically underserved groups — are especially high.[LE-LS-GL-004]

The subsections that follow address each layer of the AI/ML landscape in learning analytics: 8.4.1 covers the categories of AI techniques; 8.4.2 covers specific ML algorithms and their educational applications; 8.4.3 covers decision support frameworks that translate AI outputs into practitioner action.

¶ Subsections:

From the Learning Engineering Toolkit

The Toolkit's analytics-methods chapter sorts the available techniques into three groups — prediction, inference, and mining — and the right group depends on what the team is trying to find out. [LET-18] Predictive work trains a model to estimate one target attribute of the data: a regression approach outputs a numeric estimate derived from other input variables, whereas a classification approach assigns a categorical label by applying cutoff points to an underlying score. [LET-18]

Inference work looks instead at how an outcome relates to a set of predictors. [LET-18] One branch, sometimes called latent knowledge estimation, gauges what a learner knows or can do, while another reads log traces such as clickstream records to characterize behavior — flagging, say, a learner who is trying to game the system. [LET-18] Mining, by contrast, surfaces features, regularities, relationships, or outliers that can guide a decision or seed further study, drawing on approaches such as relationship, correlation, causal, association-rule, sequential-pattern, and network analysis. [LET-18] The book is emphatic that a statistical relationship is not evidence of cause and effect, and that only the causal variety of mining supports a claim that one factor produces another. [LET-18]

All of this rests on large-scale data — enormous collections processed by computer to expose patterns, trends, and connections, particularly in how people act and interact. [LET-06] Rich data from instrumented learning environments make it possible to spot regularities and forecast results; knowledge tracing, for instance, infers each learner's grasp of individual skills or concepts from their sequence of attempts. [LET-06] Since a raw log can contain millions of events spanning thousands of learners, reduction techniques such as dimensional reduction trim the variable count while retaining what is relevant to the question at hand. [LET-06]

The book characterizes machine learning as a branch of AI in which software improves from exposure to data via algorithms and statistical models, with a model fit on a training set, adjusted against a validation set, and finally judged on a held-out test set. [LET-06] Duolingo, for example, opened a public competition inviting outside teams to determine which algorithms most accurately anticipate the errors an individual learner will make. [LET-06] Regardless of technique, the book insists a model always be evaluated on data separate from whatever was used to build or train it. [LET-18]

Sources from the Learning Engineering Toolkit

[LET-06]Michelle Barrett, Erin Czerwinski, Jim Goodell, Daniel Jacobs, Steve Ritter, Robert Sottilare & Khanh-Phuong Thai (2023). Chapter 6: Learning Engineering Uses Data (Part 2): Analytics. In Jim Goodell & Janet Kolodner, Learning Engineering Toolkit (pp. 175–199). Routledge / Taylor & Francis. doi:10.4324/9781003276579
[LET-18]Erin Czerwinski, Tanvi Domadia, Scotty D. Craig, Jim Goodell & Steve Ritter (2023). Chapter 18: Data Analysis Tools. In Jim Goodell & Janet Kolodner, Learning Engineering Toolkit (pp. 365–377). Routledge / Taylor & Francis. doi:10.4324/9781003276579

Community Recommended Reading

Data-Informed Course Improvement: The Application of Learning Engineering in the Classroom — Anne Fensie (2023, ICLS 2023)

ID: lecommons-rl-data-informed-course-improvement-the-application-of-learning-engineeri · type: reading list item · confidence: medium · expert-validated: false

Chapter 6: LE Toolkit — Learning Engineering Applies the Learning Sciences — Jim Goodell, Janet Kolodner, Aaron Kessler (2022)

Source: goodell-ch6-le-applies-learning-sciences · confidence: medium · expert-validated: false

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

8.4 Artificial Intelligence and Machine Learning in Learning Analytics

AI in Learning Analytics: A Layered History

The Role of the Simon Initiative and Open Infrastructure

The Learning Engineering Toolkit's AI Framing

¶ Subsections:

From the Learning Engineering Toolkit

Sources from the Learning Engineering Toolkit

Further Reading

Landmark Academic Papers

Policy, Reports & Grey Literature

Organizations

Community Recommended Reading