About the model

NBAPredict estimates the probability that a team wins a given game. It is built on the same proven pipeline as our sumo and snooker sites: a temperature-scaled ELO rating blended with Glicko-2 (which tracks how certain each rating is), with a gradient-boosted correction layer planned once enough match data accumulates.

Home advantage

Each team carries an ELO+Glicko rating per competition. The home side gets a rating bonus (tuned per league); neutral-site games drop it. NBA basketball games are decisive, so the ensemble gives a clean home/away split.

Performance

The performance page will track calibration — Brier score and reliability — once enough matches are ingested to score the model honestly.

These are model probabilities for entertainment and analysis, not betting advice.