Week 5 Tutorial Plan — Facilitator Guide

AI for Prediction & Forecasting: Part 1 · 2 hours · connected classrooms (MY + AU) · LO5

Audience: students who watched the Week 5 lecture (Acts I–II) and posted a "prediction that got it wrong" to the board. Flipped model — do not re-teach the lecture.

Co-teaching: two leads. Lead A drives the room and the clock. Lead B drives Meet, breakouts, chat and cross-campus parity. Swap next week.

You need: the recap slides, activity_slides.pptx projected, the printed or shared class_activity.md worksheet, Meet with breakout rooms pre-created, and a visible timer.

Scope note. This is the judgement week. Diagnosing broken forecasts is the whole job. The build happens next week — if a team wants to start building today, point them at the Week 6 pre-class and the ten items they need to bring.

At a glance

Time	Segment	Lead	Mode
0:00–0:10	Welcome + "a prediction that broke" round	A	Whole class
0:10–0:25	Recap the two acts — interactive, not a re-lecture	A & B	Whole class
0:25–0:30	Form teams, hand out the worksheet	B	—
0:30–1:20	Forecast Forensics — four broken predictions	A & B float	Teams
1:20–1:40	Debrief + the answer key	A	Whole class
1:40–1:55	MVP check-in + Week 6 pre-flight	A & B	Teams
1:55–2:00	Close + to-do	A	Whole class

0:00–0:10 — Welcome + round (Lead A)

Fast round: one prediction each that got it wrong, from the board posts. Collect three or four out loud and hold them — you will use them in the debrief.

Set the norm: "This room runs on disagreement. If you think a diagnosis is wrong, say so."

0:10–0:25 — Recap, interactively (Lead A & B)

Pull the answers out of the room rather than replaying the lecture:

A prediction is a distribution, not a point. Ask: what is missing from "sales will be RM70,000"?
The five traps. Put them up blank and have the room fill them in.
The accuracy lie. Put up "96% accurate, catches 12% of fraud" and ask why both numbers are true at once. This is the one that has to land — Week 6 builds directly on it.

Lead B seeds the chat for the quieter campus.

0:25–0:30 — Teams and worksheet (Lead B)

Teams of 3–4, mixed-campus where rosters allow. Worksheet into every breakout chat.

0:30–1:20 — Forecast Forensics (both float)

Four broken predictions. For each: name the failure, then prescribe the fix. Rotate through rooms — Lead A the in-room groups, Lead B the Meet rooms.

Prompt, don't answer: "What would you need to see to believe this number?"

Watch for teams that spot the symptom but not the cause — "the model is bad" is not a diagnosis.

1:20–1:40 — Debrief + answer key (Lead A)

Walk the four cases, asking first and revealing second. Spend the most time on the class-imbalance case, because it is the one that recurs all semester and it is the bridge into Week 6.

Land the three honest-evaluation questions: what is the base rate · which error is worse · is it still true?

1:40–1:55 — MVP check-in and Week 6 pre-flight (both leads)

Next week is a build week and it fails if teams arrive empty-handed. Use this block to make sure every team leaves knowing:

Which of the four lanes their MVP feature is in (predict a number / flag or classify / extract from documents / retrieve and generate). The Week 6 pre-class reading explains them.
The ten real items they must bring — history, labelled cases, documents, or questions plus a corpus, depending on the lane.
That their team repo exists with an empty DIARY.md. Chase this now; a team that turns up without a repo loses twenty minutes of build time next week.

1:55–2:00 — Close

Lead B drops: the Week 5 post-class task and the Week 6 pre-class (with the "bring ten items" instruction highlighted).

Cross-campus protocol

Mixed-campus breakouts where rosters allow. Lead B owns admissions, breakouts and parity.
One instruction visible on the shared screen at all times, so a student who drops and rejoins knows what to do.
Parity check each segment: if one campus has been quiet for more than five minutes, Lead B pulls a name from there.
Hand-offs verbal and explicit.

Contingencies

Few watched the lecture. Do not re-lecture. Run the round longer and teach the five traps through the four cases — they are the content.
Running long. Cut a case rather than cutting the debrief; the diagnosis discussion is where the learning is.
Meet drops. The worksheet works on paper — keep the room running and let B re-establish the link.
Strong cohort finishing early. Have them bring a real prediction from their own project and run the same diagnosis on it. That doubles as Week 6 preparation.

BEX3202 — AI Applications in Business · Monash University · CRICOS 00008C