Stacking integration algorithm Based on Regression models and Informer for Olympic Medal Standings Forecasting

Abstract

As the most influential sports event in the world, the Olympic Games is a wind vane for evaluating a country's sports strength, resource allocation, development strategy and competitive level. With the end of the 2024 Paris Olympics and the approach of the 2028 Los Angeles Olympics, scientific prediction of the number of medals has become a topic of great interest. However, challenges such as the emergence of first-time medal-winning countries, the host country effect, and the impact of additional events complicate the prediction process. This study proposes a comprehensive medal prediction framework that integrates multiple data science methods. The study introduces a novel country classification method based on K-means++ clustering, which categorises the participating countries into sports powerhouses and emerging countries, thus greatly improving the model's adaptability and prediction accuracy. In addition, the study adopts a hybrid modelling approach that combines multiple regression models (e.g., Random Forest, XGBoost) with time series models (Informer) through stacked integration to improve robustness and generalisation. For prediction uncertainty, the framework calculates 95% confidence intervals and incorporates prediction bias analysis to validate model stability. This study provides a robust and reliable tool for Olympic medal forecasting, offering valuable insights to policy makers and sports analysts.