Machine learning for sports betting - should model selection be based on accuracy or calibration?

Aug 3, 2026

Upload by Hazel

Sports betting in the USA has expanded alongside rapid progress in machine learning, enabling bettors to use data-driven probability estimates to spot favourable bookmaker odds. The work challenges the common practice of selecting models by prediction accuracy, proposing that calibration better reflects how close predicted probabilities are to true outcome probabilities. Experiments train models on NBA data across seasons and test betting returns using published odds, showing consistently higher return on investment with calibration-based selection than accuracy-based selection. Results also support profitable Kelly betting only under well-calibrated models.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages17

Reading Time43 min

AccessFree to download

Introduction

Sports betting and machine learning background

Why accuracy may be insufficient for betting

Calibration as a model evaluation metric

Betting framework and value bets

Moneyline wagers and implied probability

Fair odds vs bookmaker margin

Graphical Abstract Highlights Machine learning for sports betting: should model selection be based on accuracy or calibration? Conor Walsh, Alok Joshi • Model calibration is more important than accuracy for sports betting • Sports bettors can increase their wealth by a third over a single season • Kelly betting only works with a well-calibrated model Machine learning for sports betting: should model selection be based on accuracy or calibration? Conor Walsha,, Alok Joshia a Department of Computer Science, University of Bath, Somerset, UK Abstract Sports betting’s recent federal legalisation in the USA coincides with the golden age of machine learning. If bettors can leverage data to reliably predict the probability of an outcome, they can recognise when the bookmaker’s odds are in their favour. As sports betting is a multi-billion dollar industry in the USA alone, identifying such opportunities could be extremely lucrative. Many researchers have applied machine learning to the sports outcome prediction problem, generally using accuracy to evaluate the performance of predictive models. We hypothesise that for the sports betting problem, model calibration is more important than accuracy. To test this hypothesis, we train models on NBA data over several seasons and run betting experiments on a single season, using published odds. We show that using calibration, rather than accuracy, as the basis for model selection leads to greater returns, on average (return on investment of +34.69% versus -35.17%) and in the best case (+36.93% versus +5.56%) . These findings suggest that for sports betting (or any probabilistic decision-making problem), calibration is a more important metric than accuracy. Sports bettors who wish to increase profits should therefore select their predictive model based on calibration, rather than accuracy. Keywords: Decision theory, Machine learning, Uncertainty, Calibration, Sports Betting 1. Introduction Sports betting in the US is conservatively estimated to be a $150 billion industry legalsportsbetting (2022) . As a result of the worldwide interest in US sports, an abundance of data is publicly available. Much research has been done on the use of machine learning (ML) for sports outcome prediction. The success of this research has turned this data into an invaluable commodity for sports bettors around the world. If a bettor can leverage data to accurately estimate the true probability of a sporting outcome, they can identify a bookmaker’s mispricing of this outcome-and whether or not there is an opportunity to make a profit. The development of a proficient predictive model could therefore prove extremely lucrative. The National Basketball Association (NBA) in North America is the world’s premier basketball league. This paper focuses on the development of a data-driven betting system in the case of the NBA. ML for sports outcome prediction has been widely studied, however very little of this research extends to sports betting. In the bulk of this work, ML models are evaluated on accuracy achieved. Accuracy is a suitable model evaluation metric for the sports outcome prediction problem, as the goal is to correctly predict the winning team Bunker and Thabtah (2019) . However this is not necessarily true for the sports betting problem, where the goal is to estimate the true probability of the sporting outcome to identify and profit from any mispricing of the odds offered on this outcome. As calibration is used to estimate how close a model’s predicted probabilities are to Email addresses: [conorwalsh206@gmail.com](conorwalsh206@gmail.com) (Conor Walsh), [aj2151@bath.ac.uk](aj2151@bath.ac.uk) (Alok Joshi) the true probabilities, we hypothesise that calibration is a more appropriate metric than accuracy for the sports betting problem, and that basing model selection on calibration, rather than accuracy, leads to greater profit generation. The simplest way to test this hypothes

FAQs

What betting implication does the study draw for Kelly betting?

Kelly betting is stated to work only with a well-calibrated model, indicating that correct probability estimates are essential for maximizing wealth in this framework.

Aug 3, 2026

Upload by Hazel