With the global focus on Olympic performances rising, predicting medal distributions and evaluating coaching strategies are vital for national sports planning. This paper builds a comprehensive modeling framework using historical Olympic data to solve these problems. A Pearson correlation-based descriptive statistical analysis shows weak feature correlations. Second, the data is trained with models like Random Forest (RF), Gradient Boosting Decision Tree (GBDT), CatBoost, and a Stacking ensemble model. After comparison, the Stacking-based model is chosen for prediction. It outperforms single algorithms in accuracy. A Random Forest - EDA analysis reveals that participating in competitive events like swimming and athletics boosts medal counts. For every 8% increase in event diversity, medals rise by 7.2% on average. Host countries can gain 10 - 15% more medals by choosing advantageous events, like France adding wrestling and handball. Next, the SHAP - EWH - TOPSIS framework quantifies the "Great Coach" effect. SHAP values show that a coach's experience and number of elite athletes trained are key in medal prediction. The Entropy Weight-modified TOPSIS model ranks coaches. Pep Guardiola and Lang Ping are at the top. Finally, sensitivity analysis validates model robustness. Medal predictions are most sensitive to the number of participating events and the coaches' athlete-training ability. The EWH-TOPSIS framework is stable with ±20% weight fluctuations, and coach rankings have 92% consistency. This paper offers practical guidance for National Olympic Committees in resource allocation, coach recruitment, and event participation strategy. The machine - learning and multi - criteria - decision - making model framework is a reliable tool for predicting Olympic success and unlocking sports potential.
© Copyright 2025 Advances in Engineering Technology Research. Alle Rechte vorbehalten.