Skip to main content
Sumo Oracle is an R-based machine learning tool that forecasts the outcome of sumo wrestling matches. By training on historical match data — wrestler weight, age, height, and win history — it applies two complementary models to make predictions: a Bayesian generalized linear model (GLM) and a neural network. The GLM achieves approximately 70% prediction accuracy, making it the stronger of the two models. Both models can be run side-by-side for comparison.

Introduction

Learn what Sumo Oracle does and how the two models work together.

Quickstart

Set up your R environment and run your first prediction in minutes.

Concepts

Understand the data format, model architecture, and evaluation approach.

Reference

Explore the full function reference for all helper utilities.

​
How it works

1

Prepare your data

Provide a CSV file with wrestler attributes for each match: weight, age, height, and number of wins. Mark completed matches with a result (0 or 1), and leave result blank for matches you want to predict.
2

Train the models

Sumo Oracle reads your CSV, separates historical matches from undecided ones, and fits both a logistic regression (GLM) and a neural network to the historical data.
3

Generate predictions

Run predict.glm() and predict() against the undecided matches. Each model outputs a directional prediction: the wrestler on the left wins, or the wrestler on the right wins.
4

Evaluate accuracy

Use the built-in train/test split and confusion matrix evaluation to measure how well each model performs on held-out data.
The GLM model consistently outperforms the neural network on this dataset. Start with GLM predictions and use the neural network as a secondary signal.

Build docs developers (and LLMs) love

Get started for freeTalk to us