How We Predict Football Matches
Poisson Distribution + Expected Goals + Injuries + Market Calibration = Smart Predictions
This guide is kept in sync with the live model β last updated: August 22, 2026
π‘ Data Sources
Live fixtures, results and team stats from 47+ leagues, refreshed daily.
Bookmaker odds for 1X2, BTTS and O/U markets, refreshed several times a day β most intensively in the final 24 hours before kick-off.
Chance quality created and conceded in recent matches β where the league provides it (top competitions).
Official absence data, available days before kick-off β unlike lineups, which arrive an hour before.
Head-to-head results weighted by recency β a meeting from last season counts far more than one from a decade ago.
Current tables plus last season's final standings β so a promoted side is never mistaken for a title contender in August.
π What We Analyze
Last 5 matches weighted by recency β newer results count more. Shown as WDL in tables.
Where available, recent xG is blended with raw goals β because chance quality predicts future scoring better than the scoreline alone.
Every confirmed injury or suspension reduces a team's attacking expectation. The more players out, the bigger the adjustment β capped so it never distorts a match on its own.
Direct encounters with a 4-year half-life: last season's meeting weighs full, a 2016 cup tie barely moves the needle.
Attack and defense stats separated by venue. Home boost and away penalty applied in the lambda calculation.
Current table position once the season settles β and last season's final standings while it's still early. Newly promoted teams start with a realistic, lower rating.
βοΈ Model + Market Calibration
Our probabilities are not just raw model output. For the Match Winner market, the final number is 75% our model and 25% the bookmaker's implied probability (with the bookmaker margin removed). This keeps extreme outliers grounded while preserving the model's own edge.
The interesting part is what remains after blending: if our probability still sits clearly above the bookmaker's implied chance, that fixture has a genuine value edge β and we show you exactly where, in the Model vs Market card inside League Insights on league pages (for example, on the Premier League predictions page).
β½ Prediction Types We Offer
What: Home win (1), Draw (X), or Away win (2).
How: We sum the full Poisson score matrix per outcome β and the pick is always the outcome with the highest total probability. No exceptions, no contradictions.
Best for: Single bets with clear favorites.
What: The match ends level.
How: A separate value market, not a runner-up. A draw tip appears only when the draw probability is high and both teams are evenly matched in strength β because a draw is never the single most likely outcome, yet lands in about a quarter of all matches.
Best for: Value hunting at longer odds (typically 3.00+) on balanced fixtures.
What: Both teams score at least 1 goal β Yes or No.
How: Sum of all score matrix cells where both sides have β₯1 goal. The pick is always the majority side of that probability.
Best for: Attack-heavy teams or weak defensive matchups.
What: Total goals Over 2.5 (3+) or Under 2.5 (0β2).
How: Sum of matrix cells with total goals β₯3 vs <3 β again, the pick always follows the majority of the probability mass.
Best for: High-scoring leagues or attacking teams.
What: Exact final score (e.g., 2-1, 1-0).
How: Highest probability cell in the full score matrix (Poisson).
Best for: High-risk/high-reward. Probability is naturally 5β15% β this is normal, not a flaw.
What: Cover two outcomes β 1X, X2, or 12.
How: Best combined probability of two outcomes from the match result model.
Best for: Lower risk. Combined probability typically 60β85%.
What: Total corners Over/Under (e.g., Over 9.5).
How: Based on team pressing style, average corners per match (home/away), and opponent tendency to defend deep.
Best for: Alternative market when match result is hard to call. Corners are less affected by individual moments like red cards or penalties.
π We Publish Our Real Accuracy
Most prediction sites tell you how good they are. We show you. Every league page carries a live AI Prediction Accuracy block: settled hit rates per market, a rolling last-30-days view, and a calibration table β when the model says 70%, you can check whether it really lands about 7 times out of 10.
The same real numbers feed the Model Track Record card in League Insights. If the model hits a cold patch in a league, you will see it there before we could hide it β that is the point.
π― Confidence vs Probability
Every prediction shows two numbers. They answer different questions β and their typical ranges differ by bet type. Understanding both is key to smarter decisions.
Raw mathematical output from the Poisson score matrix (calibrated with market odds for 1X2). The pure chance of a specific outcome.
- Match Winner: typically 30β65% β one of three outcomes
- BTTS / Over/Under: typically 45β75% β binary outcome
- Double Chance: typically 60β88% β covers two outcomes, naturally higher
- Draw tips: typically 25β35% β a draw is never the favourite, that's the nature of the market
- Correct Score: typically 5β18% β one of 30+ scorelines, always low
Normalized signal strength β the probability scaled to a 0β100 range, with a ceiling tuned per market. Alongside it we flag data quality (real, partial, pending) so you know when the model ran on thin inputs.
- 75%+ β strong signal β consider betting
- 60β74% β moderate, close match or thinner data
- Below 60% β weak signal β skip or reduce stake
Note: Double Chance and Draw confidence use their own scales β a draw tip caps lower on purpose, because no honest model should be 90% confident in a coin-flip market.
Decision Guide by Bet Type
| Bet Type | Probability | Confidence | Action |
|---|---|---|---|
| Match Winner | β₯ 55% | 75%+ | β Strong bet |
| BTTS / Over/Under | β₯ 60% | 70%+ | β Strong bet |
| Double Chance | β₯ 70% | 65%+ | β Strong bet β lower odds, lower risk |
| Draw | β₯ 27% | 70+ | β Value bet β check odds β₯ 3.20 for positive EV |
| Correct Score | β₯ 20% | 80%+ | β Strong bet β verify bookmaker odds β₯ 4.00 |
| Correct Score | 14β19% | 75%+ | π‘ Good value β odds should be β₯ 5.00 |
| Correct Score | 8β13% | 70%+ | β οΈ Acca only β odds β₯ 8.00 for positive EV |
| Correct Score | Below 8% | Any | β Skip β odds rarely justify the risk |
| Match Winner | 45β54% | 75%+ | π‘ Consider Double Chance instead |
| Any type | Any | Below 60% | β Skip or reduce stake significantly |
How to Read the Prediction Table
Every prediction table on PredictLix follows the same layout. Here is what each column means so you can act on the data instantly.
All predictions are generated exclusively through statistical and mathematical analysis of historical and current data β no human judgment, editorial input, or third-party influence is involved. The model runs entirely on numbers, including published injury and suspension data.
PredictLix predictions are designed as a decision-support tool, not a definitive signal. Use them as one input alongside your own research β last-minute lineup news, weather, derby intensity, and managerial changes are factors no statistical model can fully capture. Any final decision on scores, outcomes, or markets remains entirely at your discretion.
Predictions are provided for informational and entertainment purposes only and do not constitute betting advice. No prediction system guarantees profit. Never bet more than you can afford to lose.
This platform is operated by Timothy Omolo, Kisumu, Kenya.
π Ready to Start?
Put the system to work with today's predictions.