Context

BrightPath Education uses a binary classification model to identify students at risk of dropping out of an online certificate program so advisors can intervene early. The team reports that overall accuracy looks strong, but program directors are concerned that too many at-risk students are still being missed.

Current Performance

The Problem

The new model is more accurate and more precise than the prior framework, but recall has fallen. As a result, advisors spend less time on false alarms, yet a meaningful share of at-risk students receive no intervention.

Requirements

1. Interpret what the current metrics say about model performance and business impact.
2. Explain why accuracy alone is not sufficient for evaluating this model.
3. Compare the current model with the previous rules-based evaluation framework.
4. Identify likely reasons for the precision-recall tradeoff.
5. Recommend specific changes to the evaluation approach and model setup.

Constraints

• Advisor team can proactively contact at most 1,300 students per month.
• Missing an at-risk student is considered more costly than contacting a student unnecessarily.
• Model retraining can happen only once per academic term.