Background: Racehorses are among the most important sporting and economic animals due to their unique genetic and physiological capacities. Accurate prediction of breeding values (BVs) is critical for enhancing performance and guiding selective breeding. Traditional methods such as Best Linear Unbiased Prediction (BLUP) are accurate but computationally intensive, while machine learning (ML) offers a promising alternative for analyzing large-scale phenotypic and pedigree data.Objectives: This study aimed to systematically evaluate the predictive performance of 13 ML algorithms for estimated breeding values (EBVs) of Race Completion Time (RCT), Rank at the End of Competition (REC), and Average Speed (AS) in Iranian crossbred racehorses.Methods: The dataset included 55,070 race records from 5,296 horses. EBVs were estimated using single-trait mixed models (STM) via restricted maximum likelihood in Wombat software. ML models, including linear models, tree-based ensembles, boosting methods, and artificial neural networks (ANNs), were trained on 75% of the data and validated on 25%. Model performance was assessed using accuracy, correlation (r), mean squared error (MSE), mean absolute error (MAE), bias, and dispersion.Results: Among all models, ANN achieved the highest predictive accuracy (RCT: r = 0.72; REC: r = 0.85; AS: r = 0.70), outperforming STM and other algorithms. Tree-based ensembles and boosting methods, including XGBoost, LightGBM, Gradient Boosting, and Random Forest, also showed robust performance.Conclusions: These results highlight the potential of ML, particularly ANN and boosting approaches, to improve EBV prediction, remain robust under incomplete or low-heritability data, and enhance the efficiency and reliability of genetic selection in horse breeding programs.