Finding your feet: A Gaussian process model for estimating the abilities of batsmen in Test cricket

TL;DR

This paper introduces a Bayesian Gaussian process model to dynamically estimate and predict cricket players' batting abilities over their careers, addressing limitations of traditional averages and existing ratings.

Contribution

It develops a novel Bayesian Gaussian process model that captures both short-term and long-term variations in players' batting abilities.

Findings

Model outperforms traditional batting averages in predicting player ability.

Provides a more nuanced understanding of ability fluctuations during careers.

Enhances ability prediction compared to ICC ratings.

Abstract

In the sport of cricket, player batting ability is traditionally measured using the batting average. However, the batting average fails to measure both short-term changes in ability that occur during an innings, and long-term changes that occur between innings, due to the likes of age and experience in various match conditions. We derive and fit a Bayesian parametric model that employs a Gaussian process to measure and predict how the batting abilities of cricket players vary and fluctuate over the course of entire playing careers. The results allow us to better quantify and predict player batting ability, compared with both traditional cricket statistics, such as the batting average, and more complex models, such as the official International Cricket Council ratings.