Study on Olympic Medal Prediction Based on RF Feature Selection and BP Deep Learning

Abstract

This paper develops a model to predict the medal distribution at the 2028 Los Angeles Olympics. Combining historical data and external factors, it utilizes Random Forest (RF) for feature selection and a Backpropagation (BP) neural network for prediction. The RMSE is 0.381613, and R² is 0.61. The prediction shows that the US will lead the medal table, Canada will improve its performance, and the Netherlands may decline. It identifies potential medal-winning countries, such as Bangladesh, for the first time. Spearman correlation, K-means++ clustering, chi-square test, and PLS regression analyze the impact of host country events; sailing and other events show high correlation. Welch's t-test verifies that excellent coaches significantly improve medal performance for Japan and the US, but not for Canada. The paper suggests that the IOC optimize athlete experience, event selection, and coaching resources. The model is accurate but limited by historical data; incorporating dynamic factors is needed for improvement.