Classifying Winning Performances in International Women's Rugby Union

Area of Science:

• Sports Science
• Rugby Union Analytics
• Performance Indicators

Background:

• Previous research indicates relative performance indicators (PIs) are more effective than isolated PIs in predicting match outcomes in men's rugby union.
• This methodology has not been extensively applied to women's rugby union, highlighting a gap in performance analysis for the female athlete.

Purpose of the Study:

• To identify key performance indicators (PIs) that maximize prediction accuracy of match outcomes in women's rugby union.
• To compare the efficacy of isolated versus relative PIs in predicting match success using women's international rugby data.

Main Methods:

• Twenty-six PIs were extracted from 110 women's international rugby matches (2017-2022).
• Isolated and relative data sets were created, with relative data calculated by subtracting opposition PIs.
• Random forest classification models were employed for feature selection, importance analysis, and prediction on unseen data (2021 World Cup).

Main Results:

• Both isolated and relative models demonstrated high classification accuracy, with the relative model slightly outperforming the isolated model (78% vs. 75%).
• Reduced models showed strong predictive power, correctly classifying 100% of isolated and 96% of relative test data outcomes.
• Key significant PIs in the relative reduced model included meters made, clean breaks, missed tackles, lineouts lost, carries, and kicks from hand.

Conclusions:

• Increased relative meters made, clean breaks, carries, and kicks from hand, alongside decreased relative missed tackles and lineouts lost, are strongly associated with match success in women's rugby union.
• These findings provide valuable insights for optimizing physical and tactical preparation for women's rugby teams.
• The study supports the utility of relative performance indicators for targeted physiological research in women's rugby.