Module 1 Assessment: Problem Framing Check
Assessment ID: ML-M01-QA01 Estimated active time: 20-30 minutes Status: Draft
Part A: Short checks
Answer in one or two sentences.
- Why should an ML project not start by choosing an algorithm?
- What is the difference between a target and a practical goal?
- What is the unit of analysis?
- Why does prediction time matter?
- Give one example of a no-model option.
- What is one way a model can harm an affected person even if it helps the direct user?
Part B: Classify the task
Classify each request as prediction, description, automation, decision support, or a combination.
| Request | Classification | Reason |
|---|---|---|
| Show which module has the highest drop-off. | ||
| Predict which learner-module attempts may need mentor support. | ||
| Send a reminder after seven inactive days. | ||
| Show mentors a review list with risk notes and limits. |
Part C: Repair the problem frame
Weak frame:
Use ML to improve course completion.
Rewrite it using this structure:
At [prediction time], predict whether [unit] will [target] by [future time], so [user] can [action].
Then identify:
| Field | Answer |
|---|---|
| Target | |
| Unit of analysis | |
| Prediction time | |
| Outcome window | |
| User | |
| Action |
Part D: Baseline and risk
For your repaired frame:
- Name one simple rule baseline.
- Name one descriptive-analysis alternative.
- Name one unacceptable failure.
- Decide whether the project should continue, be revised, or stop for now.
Rubric
| Level | Evidence |
|---|---|
| Pass | The answer defines target, unit, timing, action, user, affected people, baseline, and one serious failure mode. It avoids causal overclaiming. |
| Revise | The answer has a useful goal but one or more core fields are vague, late data is used, or the action is unclear. |
| Not yet | The answer mainly says to build a model, choose an algorithm, or improve a broad outcome without defining the prediction and use context. |
Expected strong answer qualities
- Uses simple, concrete language.
- Names one prediction per row or case.
- Keeps prediction before action.
- Compares ML with a simpler option.
- Includes a reason not to build or not to use the model.
- Does not use real personal or sensitive data.
