Match Prediction Model
Team ratings
Each team has an attacking and defensive rating which gets updated following every match.
Expected Margin
Expected Margin=(Atth+Defh)−(Atta+Defa)+Venue
Updating team ratings following a match
These ratings are updated following each match based on the adjusted margin compared to the expected margin.
Expected scores
Expectedh=μ+(Atth−Defa)
Adjusted scores
Adjusted scores are a weighted average of a team’s actual score and their score had they kicked at an expected accuracy. The expected accuracy is a weighted average of the competition-wide accuracy and the team’s accuracy in the current season.
Adjusted=w⋅(6⋅G+B)+(1−w)⋅(c⋅(G+B)+(1−c)⋅(G+B))
Attacking rating
Each team’s attacking rating is updated based on their
adjusted score compared to their expected score. New attacking rating=Atth+kt(Adjustedh−Expectedh)
Defensive rating
Each team’s defensive rating is updated based their opponent’s
adjusted score compared to their opponent’s expected
score. New defensive rating=Defh−kt(Adjusteda−Expecteda)
Venue advantage
Venue advantage is based on the following four factors:
- travel distances for the two teams;
- number of matches the teams have played at the particular venue over the past five seasons;
- a constant home advantage which applies to all matches, with the exception of Grand Finals, where the designated home team is playing in its home state; and
- historical performance of the teams at the particular venue compared to expected performance.
Venue=Travel+Venue
Experience+Home Advantage+Venue Performance
Venue Experience=k⋅((Matchesh+c)w−(Matchesa+c)w)
Incorporating player information
The official AFL Player Ratings are used to supplement the Team Ratings in order to obtain an improved prediction based on the named team lineups for the two teams. Detailed information on the AFL Player Ratings can be found below.
For the purpose of this model, a player’s estimated rating for a match is calculated using a weighted average of their rating in previous matches with their latest match carrying the greatest weight. The estimated rating for all players listed in the team lineups are aggregated for each team.
Expected Marginp=ws⋅((Atth+Defh)−(Atta+Defa))+wp⋅(Player Ratingh−Player Ratinga)+Venue
Brownlow Medal Prediction Model
The Brownlow Medal prediction model uses an ordinal logistic regression model and was based on Chew The Stat’s Monte ChaRlo model methodology. The Wheelo Ratings model uses in-game statistics only which would generally not perform as well as a model which takes into account coaches and media votes.
Statistics
- Match result and margin
- AFL Player Rating points
- Kicks
- Handballs
- Total possessions and contested possessions
- Metres gained
- Centre clearances and stoppage clearances
- Total marks, contested marks and intercept marks
- Hitouts and hit outs to advantage
- Tackles
- Frees for and against
- Goals
- Goal assists
- Score involvements
- Bounces
- Spoils
- Defensive one-one-one contests won or neutralised