☕ Buy me a coffee View on GitHub

đŸ‡¯đŸ‡ĩ Sumo Honbasho Modelling

Interactive exploration of wrestler performance through Bayesian skill estimation

What is a Honbasho?

A honbasho (æœŦ場所) is an official sumo tournament held by the Japan Sumo Association. Six honbasho take place each year, each lasting 15 days with wrestlers competing daily. The tournaments are held in January (Hatsu), March (Haru), May (Natsu), July (Nagoya), September (Aki), and November (Kyushu).

This dashboard uses Bayesian modeling to estimate wrestler skill levels based on match outcomes across these tournaments, providing win probability predictions and tracking performance evolution over time.

â„šī¸

Matchup Win Probability Matrix

Win probabilities for all wrestlers in the selected basho (row wrestler vs column opponent). Click on cells to see details.

Sort by:
90%+ win probability
70-90%
50-70%
30-50%
<30%

Select a basho to view matchup probabilities

Current Rankings

Sort by:
Rank Name Official Rank Elo Rating Skill Score
Loading rankings...

Elo Progression

This chart shows the Elo rating progression of selected wrestlers over time. By default, it displays the top 10 currently active wrestlers.

Head-to-Head Comparison

About This Dashboard

This dashboard visualizes sumo wrestler performance using a Bayesian Bradley-Terry model with time-varying skills. The model estimates each wrestler's skill level across different basho (tournaments) and calculates win probabilities for matchups.

Model Details

  • Elo Rating: Calculated as 1500 + 400 × skill score. A difference of 400 Elo points indicates a 10:1 win probability.
  • Skill Evolution: Models wrestler skills as evolving over time using a Gaussian Random Walk process.
  • Weight Factor: Incorporates the effect of weight differences on match outcomes.
  • Head-to-Head History: Accounts for previous matchup results between specific wrestlers.

Data Sources

Historical match results and wrestler information from professional sumo tournaments (basho). Data includes match outcomes, wrestler rankings, weights, and kimarite (winning techniques).

How to Use

  • Matchup Matrix: View predicted win probabilities for all wrestlers in a selected basho. Click cells to see detailed head-to-head statistics.
  • Current Rankings: Browse the top 50 wrestlers sorted by Elo rating or official rank.
  • Elo Progression: Track how wrestler skills evolve over time. The top 10 are selected by default, but you can choose any combination.
  • Head-to-Head: Compare two wrestlers directly, including their historical matchups and predicted win probability.

Limitations

  • The model does not account for injuries, sudden changes in wrestler form, or external factors affecting match outcomes.
  • Predictions are based on historical data and may not always reflect current conditions or emerging talents.
  • Data quality and completeness can affect the accuracy of skill estimates and predictions.
  • Win probabilities are estimates and should be interpreted as such, not certainties.