I_PREDICT_SPORT
About
I_PREDICT_SPORT publishes daily match probabilities across ATP, ATP Challenger and WTA from a purpose-built prediction engine: chronological Elo-family ratings (surface-specific, margin-of-victory aware, uncertainty-scaled), an exact point-by-point serve/return match model, and gradient-boosted ensembles — trained on tour-level results and enriched with Challenger and lower-tour data that most public models never see. The women's board runs a separate model built the same way on 40,935 WTA matches.
On strictly time-ordered validation the men's model holds roughly 66% winner accuracy with a Brier score near 0.213. The women's model was measured against a bookmaker on 4,077 out-of-sample 2026 matches and is at parity overall (0.1963 against 0.1965) and ahead on grass (0.1977 against 0.2053); it is not ahead on hard courts, where the market is sharpest.
Every published probability is logged with the price at the time and audited against the settled outcome. What that audit says today is worth stating plainly: on an unselected population we do not beat a tradeable price. The board is a research instrument and a public track record, not a tipping service — it is paper throughout, and the honest negative results are published alongside the positive ones.
The person behind it
Alex Houck is a machine learning and AI engineer who builds prediction systems where being wrong has a price: sports outcome modeling, medical-device AI, surgical robotics, clinical outcome prediction, financial markets, and applied data science. His work spans the full arc from raw data pipelines to validated, deployed models with live accountability for their accuracy.
Alex has consulted for Fortune 500 companies and early-stage startups alike, and is open to consulting engagements — particularly where rigorous prediction, calibration, and honest evaluation matter more than hype.