FPL Axiom
FPL Axiom is a fully automated quantitative analytics platform for Premier League and Fantasy Premier League analysis. The core insight: most football analysis is gut feel. This platform uses Poisson distribution modeling, Bühlmann credibility theory, and z-score statistical testing to identify which teams are genuinely over- or underperforming relative to their expected goal statistics, and how likely they are to revert.
Results & Impact
The Challenge
No accessible tool existed for FPL managers to apply rigorous statistical modeling to performance data. FBRef (the primary source of Opta xG data) has no public API, and official data feeds cost thousands per month. Most analytics tools use raw stats rather than expected value models, making it hard to separate luck from skill.
The Solution
Built a fully automated Python pipeline (scraper → calculator → analyzer → visualizer → reporter) that scrapes FBRef daily via GitHub Actions using residential proxies, runs Poisson distribution models to compute Expected Points from xG, applies Bühlmann Credibility for Bayesian skill-vs-luck adjustment, calculates VAPM (Value Above Positional Mean) for FPL player ranking, and auto-commits output CSVs to trigger a zero-config Vercel redeploy of the Next.js dashboard. Uses exponential weighting (λ=0.85) for recency bias and Focal Loss to prioritise hard edge cases.