Modeling and analysis of Olympic medal table based on multiple features

Predicting Olympic medal counts holds significant value for strategic planning by National Olympic Committees. This paper presents a comprehensive modeling approach for predicting medal distribution across countries at future Olympic Games. We developed a stacked integration learning framework incorporating multiple machine learning algorithms: LR, LASSO, SVM, and Catboost in the first layer as base learners, and RF, XGBoost, and LightGBM in the second layer as meta-learners. This approach effectively quantifies medal fluctuations based on multiple country characteristics and their correlation patterns. The model was used to predict medal standings for the 2028 Los Angeles Olympics and identify countries likely to improve or decline in performance. Additionally, we constructed specialized evaluation models for countries without previous medals by parameterizing scarce data and adjusting association ratios. To analyze the relationship between sports types and national medal counts, we employed a neural network algorithm that determined sport-specific impact parameters for each country. Furthermore, we examined the “great coach” effect on medal stability using the DBSCAN clustering algorithm and UMAP hierarchical analysis, identifying optimal coach-sport pairings for six selected countries. Our modeling approach provides valuable insights for National Olympic Committees to optimize resource allocation and strategic planning for future Olympic competitions.