A Medal Prediction Model Based on XGBoost-Logistic Regression and Quantification of the “Great Coach” Effect

Abstract

The paper selected sports that have consistently appeared in the Olympic Games since 2000, created a feature system, considering factors such as historical data, the elite athletes, gender ratios, and hosting country in the games. By comparing the performance of different models in this input feature the XGBoost performed exceptionally well in forecasting with its highest R² value which indicates that it is good at explaining the fluctuations in the data and the lowest MSE value which reflect that it has less variance in its predicted values. Furthermore, paper aimed to quantify the contribution of "great coach" on the number of medals won. Using the coaching information, the paper reconstructe the feature system and applied Ridge Regression to assess the impact of coaches. The effect of Coach Lang Ping on the Chinese and American women’s volleyball teams was quantified as 6.432 and 6.913, respectively. Similarly, Coach Béla Károlyi contribution to Romanian and American gymnastics medals was 4.6123 and 4.3333, respectively. Then employed the CUSUM method to identify key breakpoints in the medal sequence and identified 3 countries (Italy and United States) that could benefit from employing “great coaches.” The results highlighted the host country and elite athletes as particularly influential factors, offering valuable insights for national Olympic committees in their decision-making processes.