Predictive machine learning models for match outcomes in taekwondo based on competitive history.

This article focuses on predicting match outcomes in Olympic taekwondo using machine learning (ML) and deep learning (DL) models based on structured historical competition data from the Peruvian Taekwondo Sports Federation. Two dataset versions were developed: one modeling individual match sequences and another capturing pairwise athlete confrontations, enabling evaluation under temporal and head-to-head frameworks. Among eight tested models, LightGBM achieved the highest F1-score (84.00%) on the sequence-based dataset, while XGBoost performed best (75.00%) on the pairwise dataset. Feature importance analyses consistently identified second-round clean points and penalties as key predictors of match results, highlighting the tactical significance of mid-fight actions. The study demonstrates the potential of ML approaches to support athlete evaluation, training optimization, and strategic planning in taekwondo, while noting limitations related to dataset size and the need for ethical considerations in model deployment.