▌ epl-betting-ai.md
▒ PATH:
▒ SIZE: 1.4 KB
▒ MODIFIED: 2026-05-23 01:29
MemPalace Archive/projects/epl-betting-ai.md▒ SIZE: 1.4 KB
▒ MODIFIED: 2026-05-23 01:29
# EPL Betting AI
## What
Poisson model for English Premier League match predictions. Independent picks — not copying tipsters. Focus on draw value.
## Status
Active — Sunday match predictions (final day of season May 24, 2026)
## Data
- GitHub: github.com/ncik8/betting-ai
- Location: `/Volumes/My Shared Files/UTM/betting-ai-repo/data/`
- ~25 seasons: E0_9596 through E0_2324
- 106 columns: results, xG stats, bookmaker odds, corners, cards
## How It Works
1. Load last 3 seasons of data
2. Calculate team home/away xG (goals scored/conceded averages)
3. Expected goals = (home_team_home_xG + away_team_away_GA) / 2
4. Poisson distribution for score probabilities
5. Compare model probabilities vs Polymarket odds to find value
## Key Insight
Polymarket consistently underprices draws by 8-12 percentage points vs historical average. Model finds value on draw in most games.
## Polymarket API
- gamma-api.polymarket.com for live odds
- CLOB (clob.polymarket.com) has historical settled markets only (2022-23 onwards)
## Draw Scorelines
Model focuses on 0-0 and 1-1 draw scorelines specifically.
Sunderland vs Chelsea: 0-0 (8.5%), 1-1 (12.7%)
Burnley vs Wolves: 0-0 (8.2%), 1-1 (12.7%)
Crystal Palace vs Arsenal: 1-1 (12.4%)
## Related
- HKJC Betting AI: `/Volumes/My Shared Files/UTM/hkjc-betting/` (horse racing, separate)
- Both use Poisson-based approach
- Betting approach: truly independent from external tipsters