Leveraging Ensemble Learning for Accurate Medal Prediction in Olympic Competitions

This study aims to address the persistent challenges in Olympic medal prediction and develop an innovative integrated learning framework that overcomes the limitations of traditional single-model approaches. We propose a layered stacked integration architecture that combines three gradient-enhanced models (LightGBM, XGBoost, CatBoost) with an underlying trio of models with different feature processing capabilities. They are synergistically integrated by a multi-layer perceptron meta-learner. The framework uses the gradient based sampling efficiency of LightGBM, the exact greedy algorithm of XGBoost for sparse features and the ordered enhancement algorithm of CatBoost for categorical variables. The neural network meta-learner performs non-linear weight adjustment via backpropagation. Our work has been validated by reliable experiments showing that it is effective and efficient. It effectively captures higher-order interactions between social and economic indicators, historical performance data and sport-specific variables, providing a new methodological paradigm for the analysis of competitive sport.