How accurate are the prediction models?
Over a large number of predictions, a model that has not found meaningful patterns
will be randomly guessing outcomes, resulting in an approximately 50% accuracy. As of the end of the 2024 NRL season, over 944 match predictions have been generated and 603 were correct predictions - a 63.88% success rate. See the model accuracy each year below. Each year there are several factors that can temporarily influence prediction accuracy such as major rule changes, new teams & significant shifts in team performance from year to yearThe prediction process
-
The historical data for sport consists of thousands of data points collected from past matches of the respective sport.
The data points consist of a combination of factors that might explain the outcome of a game, and the outcome of the game.
-
Machine learning algorithms are trained on the data to produce models capable of predicting the outcome of a game given the matchup.
The algorithm learns patterns that are often not able to be identified from the human brain due to the large
number of data points and factors considered.
-
The predictions are auto generated using the trained machine learning algorithms.
The outcome of team 1 vs team 2 can be thought to be generated by the weighted factors of historical outcomes and patterns of factors between similar matches between similar teams.
-
The outcomes of recent games are added to the dataset for the machine learning algorithms to train on for future games, learning any new patterns occuring.
What factors are considered?
A combination of key factors and the match outcomes are recorded as part of the historical dataset. The factors are extremely important and can change over time. Some example basic factors that are generic across different sports: Team or player form.
Team or player record against the opposing team or player.
Team or player record at particular locations.
Team or player record under particular conditions.