How it works
Our prediction system combines machine learning, AI review, and continuous improvement.
ML Ensemble
Three models — Random Forest, XGBoost, and Logistic Regression — trained on historical matches with real features: team form, goals avg, H2H records, bookmaker odds, shots, possession, and clean sheets. Each model votes and the predictions are averaged.
Groq AI Review
The Llama 3.3 70B model on Groq reviews the ML output, corrects probabilities based on context (injuries, form, tactics), and adds extended markets: corners, cards, exact score, double chance, and first half result.
Live Predictions
During matches, our Poisson engine recalculates all markets in real time using live stats: win probabilities, goals, corners, cards, momentum, and next goal — updated every 15 seconds.
Feedback Loop
After each match, we compare predictions vs. actual results using Brier scores. This feedback drives model retraining and continuous improvement.
Pre-match markets (15 total)
Expected Goals (xG)
A statistical measure of the quality of chances created. An xG of 1.5 means the team is expected to score ~1.5 goals based on the chances they get.
1X2 + Double Chance
Home win (1), Draw (X), Away win (2) probabilities — plus 1X and X2 double chance markets. ML generates the core 1X2, Groq adds double chance.
Goals: Over 1.5 / 2.5 / 3.5
Probability of 2+, 3+, or 4+ total goals. Includes BTTS (both teams score) and predicted exact score.
Corners & Cards
Predicted total corners (Over 8.5, Over 9.5) and cards (Over 2.5, Over 3.5). Generated by Groq pre-match and recalculated live via Poisson.
Value Bets
When our predicted probability is higher than what bookmaker odds imply, that's a value bet. The +X% edge shows how much our model disagrees with the market.
Live Markets
During matches: win probabilities, goals (Over 1.5/2.5/3.5), BTTS, corners, cards, next goal probability, momentum index, live xG, and projected final score — all recalculated every 15 seconds.