Unit 3 project: A report card for a model

All year, models have been given scores by the people who built them. Builders are not the best judges of their own work — you saw that at the exhibition.

Today your group is the inspector. You will evaluate a model you did not build, write its report card, and give a verdict: use it, fix it first, or don't use it. Your verdict must follow from your numbers.

Work in your project groups of 3–4. One period. Everything goes in report-card.odt.

Pick one:

• A. Another group's Unit 2 model. Swap with a group your teacher names. Ask them for their kept-back test examples — and, if you can, collect five more of your own.
• B. The school-bus model. datasets/bus-predictions.csv holds 50 school days: which route (Town or Village), whether the bus was really late, and what the model predicted. Families use it to decide when to leave home. Positive class: late.

1. Confusion matrix. Choose and write down the positive class. Build the matrix — actual down the side, predicted across the top.
2. Metrics. Accuracy, precision, recall, F1. Show each calculation.
3. The metric that matters. For this model's job, which mistake does more harm — a false alarm or a miss? So which metric should it be judged on?
4. Group by group. Split the results by at least one group: route, orchard, lighting, whose photos. Is it fair across them?

Fill the model card: what it does · what it learned from · how you tested it · results overall and by group · known weaknesses · who is accountable. Then the verdict:

Verdict: use it / fix it first / don't use it — because … (one sentence, with a number in it)

Each person writes what they would say to one person who relies on this model — a parent waiting for the bus, a farmer, a shopkeeper — in two sentences they would understand. Sign with your roll number. This part is marked for you alone.