A Data-Driven Approach to Predicting Olympic Medal Distribution: Integrating Machine Learning and Graph Theory

This study proposes a comprehensive medal prediction model for the 2028 Summer Olympics, providing valuable insights for the Olympic Committee. To achieve this, we conduct rigorous data preprocessing and analysis, employing K-means clustering to classify countries into distinct groups based on key attributes. We introduce an innovative evaluation framework that quantifies national competitiveness using weighted scores, forming the foundation for our medal prediction model, which integrates regression analysis and the Informer time-series model. A key focus of our research is to explore the relationship between sporting events and national medal counts by comparing Spearman correlation coefficients, while also empirically validating the host nation advantage. For medal table prediction, we implement a stacked ensemble model, combining linear regression, random forest, support vector regression (SVR), K-nearest neighbors (KNN), and XGBoost, ensuring robustness and accuracy. To address the first-time winning country problem, we reformulate it as a binary classification task using logistic regression, evaluating performance through accuracy metrics and confusion matrix analysis. Additionally, we investigate the great coach effect by modeling it as a maximum flow problem in graph theory, proving its existence via bottleneck capacity constraints. Furthermore, we conduct uncertainty quantification and hyperparameter tuning to enhance the models reliability and predictive performance. Our findings contribute to a data-driven understanding of Olympic medal distributions, offering a novel perspective on factors influencing national athletic success.