Skip to contents

bouncer

Cricket analytics R package with a 5-component prediction pipeline featuring ball-by-ball skill tracking, match simulation, and pre-match predictions.

Overview

bouncer tracks player, team, and venue skill at the ball-by-ball level using residual-based indices built on top of XGBoost context models. It combines a 3-Way ELO system (batter + bowler + venue) with PageRank network quality adjustments to power match simulations and predictions across T20, ODI, and Test formats.

 AGNOSTIC MODEL ──> SKILL INDICES ──> FULL MODEL ──> SIMULATIONS
       │                  │                │               │
       └── baseline ──────┴── residuals ───┘     PRE-GAME PREDICTIONS

Installation 

# Install from CRAN
install.packages("bouncer")

# Or install the development version from GitHub
remotes::install_github("peteowen1/bouncer")

Quick Start 

library(bouncer)

# Step 1: Install pre-processed cricket data (one-time setup, ~500 MB)
install_all_bouncer_data()

# Step 2: Look up a player
get_player("V Kohli", format = "t20")

# Step 3: Detailed player analysis (batting, bowling, skill history)
analyze_player("JJ Bumrah", format = "t20")

# Step 4: Pre-match win probability
predict_match(
  team1 = "India",
  team2 = "Australia",
  venue = "MCG",
  match_type = "T20"
)

# Step 5: Ball-by-ball match simulation
quick_match_simulation(format = "t20")

# Step 6: Visualize win probability over a completed match
plot_win_probability(match_id = "1415755")

Core Functions

Player Analysis

Team Analysis

Predictions & Simulation

Visualization

Venue & Match Analysis

Data Management

See the full function reference for all 120+ exported functions.

How It Works

The 5-Component Pipeline

  1. Agnostic Model - An XGBoost model predicts delivery outcomes (runs/wickets) using only context features (over, wickets fallen, match phase) with no player/team/venue identity. This establishes a baseline expectation.

  2. Skill Indices - Residual-based indices track how players, teams, and venues deviate from the agnostic baseline. Updated ball-by-ball using exponential moving averages:

    residual = actual_outcome - agnostic_expected
new_skill = (1 - alpha) * old_skill + alpha * residual

  3. Full Model - A second XGBoost model uses all agnostic features PLUS skill indices to make maximum-accuracy delivery predictions.

  4. Simulations - Ball-by-ball match simulation using the full model, drawing from the predicted outcome distribution (wicket, 0, 1, 2, 3, 4, 6).

  5. Pre-Game Predictions - Combines team ELO, aggregated player skills, venue characteristics, form, and head-to-head records to predict match outcomes before the first ball.

3-Way ELO System

The core rating system maintains separate ELO ratings for three entities on every delivery:

Entity Weight What it captures
Batter 52% Individual batting ability
Bowler 22% Individual bowling ability
Venue (session) 21% Short-term pitch conditions (prep, dew, deterioration)
Venue (permanent) 5% Long-term ground characteristics (size, typical pitch)

PageRank Quality Adjustment

A network centrality system detects when players accumulate high ratings by playing in isolated, low-quality competitions. Players who face strong opponents in well-connected tournaments receive quality-adjusted ratings.

Data Source

All data comes from Cricsheet, stored in a local DuckDB database. The package handles downloading, parsing, and database management automatically.

Vignettes

License

MIT

Acknowledgments

  • Data: Cricsheet
  • Built with: DuckDB, XGBoost, dplyr, ggplot2

Citation

If you use bouncer in research or publications, please cite:

@software{bouncer2026,
  title = {bouncer: Cricket Analytics with Ball-by-Ball Skill Tracking},
  author = {Pete Owen},
  year = {2026},
  url = {https://github.com/peteowen1/bouncer}
}