Olympic Medal Prediction via Hybrid Time-Series and Static Feature Modeling

As the Olympic Games serve as the pinnacle of global sports competition, accurate medal prediction not only aids in formulating national sports development strategies but also helps identify key factors influencing athletic performance. Traditional medal prediction methods mainly rely on historical data or expert judgment, but these approaches struggle to incorporate dynamic factors such as economic conditions, athlete reserves, and host country advantages, limiting their predictive accuracy. Additionally, challenges such as the cold start problem for emerging countries further complicate prediction tasks. To address these issues, we propose an Olympic medal prediction model that integrates both time-series and static features. Our ensemble model combines LSTM, which captures long-term temporal dependencies in medal trends, with XGBoost, which performs nonlinear regression on static Olympic-year features such as GDP, population, age-eligible athletes, and host country influence. The model mitigates cold start issues by leveraging economic and demographic proxies to estimate potential medal performance for countries with limited historical records. Experimental results demonstrate that the XGBoost-LSTM model outperforms multiple baseline methods, significantly improving prediction accuracy and validating the effectiveness of integrating time-series and static features. Through this study, we further reveal the impact of economic strength, athlete development, and geographical factors on Olympic medal distribution, providing new insights and methodologies for future Olympic performance forecasting.