AFL
Prediction Model

Untouched 2022 test season

75.2%tips correct
25.34point margin MAE
207matches tested

This week

Semi-Finals

Probabilities are for regulation time; a draw in finals leads to extra time.

Perth Stadium
F FremantleHome 48.4%
G GeelongAway 50.4%
Model call Geelong by 1 1.2% draw · 80% range: Geelong by 43 to Fremantle by 41
Gabba
BL Brisbane LionsHome 64.2%
A AdelaideAway 34.7%
Model call Brisbane Lions by 12 1.1% draw · 80% range: Adelaide by 30 to Brisbane Lions by 55

How predictions are made

Key prediction indicators

The 2022 test impact is the extra average margin error after an indicator was shuffled. Higher means it was more useful to the model, not that it directly adds points.

Indicator 2022 test impact
Longer-term margin form

Scoring margin from earlier matches, updated slowly so roughly the last five to ten games describe sustained form rather than one result.

+1.5 points extra average error after shuffling
Team strength rating

An Elo-style rating that starts each club near 1500, adjusts for opponent strength and result margin, and moves one-third toward average each season.

+1.16 points extra average error after shuffling
Scoring shots

Recent goals plus behinds. It measures how often a team creates a scoring opportunity, regardless of whether those shots were converted.

+0.57 points extra average error after shuffling
Expected-score defence

The expected score conceded from recent opposition chances. Lower values indicate that a team is allowing fewer or lower-quality opportunities.

+0.42 points extra average error after shuffling
Inside 50s, longer term

Forward-50 entries across prior matches, updated slowly to capture sustained territory and chance creation.

+0.23 points extra average error after shuffling
Estimated player strength

Rolling AFL player-rating output for each club's most recent lineup, with limited samples pulled toward a player's established rating.

+0.22 points extra average error after shuffling
Recent versus longer-term form

Recent form gives the newest match 30% weight. Longer-term form gives it 12%.

Match-level margin influence

Open “Why this prediction?” to see each leading signal's estimated margin influence. These are model effects, not causal claims.