About
Machine learning meets competitive League of Legends. We analyze professional match data to deliver real-time win probability predictions.
What We Do
Riftcast is a data-driven platform that forecasts the outcomes of professional League of Legends matches. We track teams across major leagues worldwide, build statistical profiles from their performance history, and run trained machine learning models to generate win probabilities before each match.
Whether you're a fan looking for informed insights or an analyst exploring data trends, our models provide a quantitative edge grounded in real match statistics — not guesswork.
Riftcast covers all major professional leagues including LCK, LPL, LEC, and LCS, as well as regional leagues like LFL, Prime League, TCL, CBLOL, and more. Our system processes thousands of individual game records to build a comprehensive picture of the competitive landscape.
Our Models
Riftcast uses three distinct machine learning models, each with different strengths. Every model is retrained daily on the latest match data to stay current with meta shifts and roster changes.
FastTree
A gradient-boosted decision tree algorithm that serves as our baseline model. FastTree provides consistent, reliable predictions across all match types. It processes team Elo ratings, historical performance metrics, and rolling averages to produce win probabilities for any matchup.
LightGBM
A more advanced gradient boosting framework that adapts to meta shifts by tracking team performance across patches. LightGBM is particularly effective at capturing the impact of international competition and playoff pressure on team performance.
PCA Sweep
Tested across 40,000+ different setups before picking this one. While the other models weigh each team stat on its own, this one looks for groups of stats that move together — like "this team scales late" or "this team snowballs early" — and predicts based on those bigger patterns.
Each model comes in two variants: a base version that predicts from team strength alone, and a draft-aware version that factors in champion selections, role matchups, team synergy, counter-picks, and player-specific champion proficiency. During live games, Riftcast automatically generates draft predictions once both teams have locked in their compositions.
Data and Coverage
Our database contains thousands of professional matches spanning multiple seasons and dozens of leagues. For every match, we store detailed statistics including kills, deaths, assists, gold income, damage dealt, vision score, objective control, and more — broken down by individual player and by team.
Data is collected automatically from multiple sources throughout the day. New match results are scraped within hours of completion, and our models retrain every morning at 06:00 CET to incorporate the latest results. During live broadcasts, Riftcast also tracks real-time game state including kills, gold, towers, dragons, and barons.
Team Elo ratings are computed chronologically — each rating reflects only the information available at that point in time, preventing any future data leakage. This ensures that predictions are fair and reflect what could genuinely be known before a match starts.
How It Works
Data Collection
We continuously gather match results, player statistics, draft compositions, and performance metrics from professional leagues around the world. This data forms the foundation of every prediction.
Model Training
Our ML models learn from hundreds of historical matches using features like Elo ratings, team performance stats, rolling averages, and champion draft analysis. Multiple algorithms compete to find the best fit.
Prediction
Select any two teams and get an instant win probability. Add a champion draft for even more precise estimates. Our models update daily as new match data comes in.
Transparency and Accuracy
We believe in full transparency about our model performance. Every prediction is stored and evaluated after the match completes, and we publicly display each model's accuracy, AUC (Area Under the ROC Curve), and recent prediction history directly on the homepage. You can see exactly how often each model gets it right.
Model metrics are recalculated after each training cycle and displayed on the homepage under Model Performance. Historical predictions for recent matches are shown alongside actual results so you can judge model quality for yourself.
Frequently Asked Questions
How often are predictions updated?
Models are retrained every morning using all available match data. Predictions for upcoming matches are generated automatically after training completes and whenever the match schedule changes. During live games, draft-aware predictions are generated as soon as champion select finishes.
Which leagues do you cover?
We cover all major professional leagues tracked by LoL Esports, including LCK, LPL, LEC, LCS, LCP (PCS), CBLOL, TCL, and many regional leagues such as LFL, Prime League, LES, First Stand, and EMEA Masters. Any league with publicly available match data and an esports schedule can be tracked.
What data goes into a prediction?
Each prediction combines dozens of features: team Elo ratings, overall win rates, average gold and XP differentials at key time marks, first objective rates, vision control, kill participation, damage output, and rolling performance averages over the last 5 and 15 matches. Draft-aware predictions additionally factor in champion meta scores, player mastery, team synergy, and counter-pick advantages for each role.
Why do different models give different probabilities?
Each model uses a different algorithm with different strengths. FastTree is a stable baseline, LightGBM is better at capturing meta trends, and PCA Sweep uses advanced feature reduction to find hidden patterns. Disagreement between models is normal and can itself be informative — when all three agree, the prediction tends to be more reliable.
Is Riftcast free to use?
Yes, Riftcast is completely free. All predictions, live game tracking, team rankings, and match history are available to everyone. We support the platform through non-intrusive advertising.
Disclaimer
Riftcast predictions are provided for entertainment and informational purposes only. While our models are built on real match data and tested methodology, no prediction system can guarantee outcomes in competitive esports. Past performance does not guarantee future results.
Riftcast is not affiliated with or endorsed by Riot Games, Inc. League of Legends and all related properties are trademarks or registered trademarks of Riot Games, Inc.
Contact
Have questions, feedback, or want to report an issue? Reach out at:
riftcast.support@gmail.com