A HYBRID SEQ2SEQ AND BAYESIAN OPTIMIZATION FRAMEWORK FOR PREDICTING OLYMPIC MEDAL DISTRIBUTION WITH UNCERTAINTY ANALYSIS

This paper proposes a robust and data-driven methodology to forecast Olympic medal outcomes with high accuracy and interpretability. A sequence-to-sequence (Seq2Seq) neural architecture is employed to learn temporal dependencies in national Olympic performance, while hyperparameter optimization is conducted using a Tree-structured Parzen Estimator (TPE) to enhance model generalization. To ensure data integrity, preprocessing steps include structured data cleansing and the use of a backpropagation neural network to address missing values. The model further integrates features such as national investment in sports, historical medal trends, and host country effects. In addition to deterministic predictions, uncertainty is quantified through Monte Carlo sampling and confidence intervals, providing probabilistic insights into future outcomes. Experimental results show that the proposed approach outperforms baseline models, achieving an R 2 improvement from 0.827 to 0.875 on the test dataset. The framework is applied to predict the medal distribution for the 2028 Los Angeles Olympics and highlights emerging medal-winning countries. These findings demonstrate the framework’s potential to assist national committees and policy makers in strategic planning for future Olympic participation.