Activity: Create an Algorithm Candidate Plan

Estimated active time: 45-70 minutes

Scenario

A fictional learning platform wants to predict whether a learner may need extra support in the next two weeks. The data is synthetic and not from real learners.

You have:

• course progress percentage;
• recent practice activity;
• number of assessment attempts;
• days since last activity;
• course level;
• support-ticket count;
• prior course completion;
• a binary label showing whether support was needed later.

Use:

• data/learner_success_algorithm_v1.csv
• notebooks/module-01-planning-notebook.ipynb
• learner-resources/algorithm-comparison-report-template.md

Task

Establish the comparison frame every later module reports against. Fit both baselines, record them, and write the selection criteria you will hold each algorithm to.

Required sections

1. The decision the model supports, and the unit of prediction.
2. Majority-class baseline: the accuracy of always predicting the most common label.
3. Logistic regression baseline, on the same split.
4. The metric you will use throughout, and why it suits this target.
5. The margin over baseline an algorithm must clear to earn consideration.
6. Limitation note.

Rules

• Fix the split now and reuse it in every later module, or the comparisons are not comparable.
• Report the majority-class figure even when it is embarrassing — it bounds every later claim.
• Do not tune anything in this module. This is the reference point.

Output

Use learner-resources/algorithm-comparison-report-template.md or your own one-page memo with the same sections.