How It Works

Understanding the data is key to making smarter decisions. Here’s a breakdown of how our tool works.

🧠 Our Dual-Model System

Our platform is powered by two distinct machine learning models working in tandem to provide nuanced insights.

1. The Predictor Model: This model is trained on years of historical AFL data to forecast a player's potential disposals in an upcoming match.
2. The Confidence Model: This second model analyzes the initial prediction and calculates a confidence score based on factors like player consistency, form, and opponent strength. This adds a crucial layer of reliability to each forecast.

⚠️ Current Limitations

While our models are powerful, they are not omniscient. It's important to understand what is not currently factored into the predictions. Complex, real-world events like last-minute player position changes (e.g., a key midfielder moving to a half-back flank) or the cascading effects of a key player getting injured mid-game are not yet fully incorporated. We are continuously working to improve the models' context awareness.

📊 Reading the Predictions Table

Each row in the Player Predictions table gives you a detailed statistical snapshot. Here's what the key columns mean:

15+, 20+, 25+: The percentage probability that the player will get *at least* that many disposals. Higher percentages signal stronger confidence in hitting those milestones.

Conf. Within (±3, ±5): The percentage probability that the player's final disposal count will be *within* that range of the prediction. For example, if a player's prediction is 22, the "±3" column shows the chance of them finishing with 19 to 25 disposals.

💰 How We Find Value Bets

A "value bet" occurs when our model believes the likelihood of an outcome is greater than the probability implied by a bookmaker's odds. We calculate the bookie's implied probability from their odds and compare it to our model's probability. If our model gives a player a 60% chance to achieve a milestone, but the bookie's odds imply only a 50% chance, we flag it as a positive Expected Value (EV) bet.

✨ The Same Game Multi

For most games, our program will suggest a Same Game Multi. It's important to note that this multi is **not** strictly a combination of the highest-value legs. Instead, it's designed to be a more balanced and reasonable option for the casual bettor, typically with final odds between $1.80 and $3.50.

📈 Data & Update Schedule

Our initial player predictions depend on third-party data, so availability for a new round can vary. Generally, main predictions are generated and available from **Tuesday morning** onwards.

All odds-based information, including **Value Bets**, the **Same Game Multi**, and **True Parlays**, is refreshed automatically on the following schedule (Perth time, AWST):

💸 Free to Use

This platform is currently free for everyone to use and try out. Our main focus is on improving the models and adding valuable features. As the tool gets better and depeding on operational costs, we may explore premium options in the future, but for now, we're excited to offer it to the community for free.

📬 Suggestions & Feedback

Spot a bug? Have a suggestion? You can reach me at:aflml.project@gmail.com

⚠️ Disclaimer

This project is for informational and entertainment purposes only. There is no guarantee of accuracy or profitability. Please gamble responsibly and only wager what you can afford to lose.