Syllabus

Meetings

Tuesdays and Thursdays, 4:15–5:30 p.m., Spring 2027. Class meetings are primarily lecture. Attendance is expected and does not affect the grade.

Purpose

Students should leave with the technical skill and critical judgment to shape how machine learning is used in education. Two to three years on, that means applying these methods to educational problems, building and evaluating models, and deciding when a model supports learning and when another approach fits better. The course asks students to move between technical methods and educational questions, including whether a result shows a benefit for learners.

Learning goals

• Define a prediction task and identify information available at the intended prediction time.
• Build a reproducible workflow with preprocessing, validation, tuning, and held-out evaluation.
• Compare regularized, kernel, tree-based, and neural models using metrics appropriate to the outcome.
• Apply knowledge tracing and text embeddings to selected educational questions, and report the limits of the evidence.
• Examine uncertainty and performance across student groups.
• Explain the distinction between prediction and evidence about an intervention.
• Communicate an educational analysis through reproducible code and a research paper.

Software

Students are free to use either R or Python for assignments and the course project. The same questions and grading criteria apply to both languages. Required work does not depend on paid model APIs or specialist GPU hardware. Text and embedding examples can use open models, modest datasets, or precomputed embeddings.

Course work

Take-home assignments and the project count for roughly equal shares of the grade. There is no final exam. Grades use a straight scale. There is no extra credit and no revision for a higher grade.

Problem sets are individual. The project may be completed by one student or by a group of two or three. A group submission includes a contribution statement naming what each member did.

Course work also includes in-class modeling exercises and project check-ins.

Late work

Each unexcused day an assignment is late lowers that assignment’s grade by 5 percentage points. A submission one minute after the deadline loses 5 points. Each further 24 hours loses another 5. An assignment that would have earned 97, submitted just after the deadline, earns 92. One submitted 26 hours late that would have earned 90 earns 80.

Request an excused late day before the deadline by emailing [email protected]. If the request is approved, the penalty does not apply. Medical and personal health issues, and other extenuating circumstances, are the reasons that qualify.

In a medical emergency, take care of your health first. A medical extension can be requested after the deadline with a note from a care provider.

Readings

An Introduction to Statistical Learning is the foundational text, in the R or Python edition. Relevant chapters cover linear regression, classification, resampling, regularization, tree-based methods, support vector machines, and deep learning. Tidy Modeling with R provides additional R workflow examples.

Big Data and Education, by Ryan S. Baker (2025, 9th edition, University of Pennsylvania), is a complementary open textbook. Selected videos on validation, feature engineering, knowledge tracing, and embeddings are linked from the topics page.

Course communication

Enrolled students will use Brightspace for announcements, submissions, grades, and feedback. This website provides selected public learning materials.

Instructor

Conrad Borchers, Vanderbilt University. [email protected].