Research on Olympic Medal Table Prediction Based on Graph Convolutional Neural Network

Abstract

As a global top sports event, Olympic medal table prediction is of great significance to the development of national sports strategies. In this study, we break through the limitations of traditional regression analysis by integrating historical medal data, host country effects, athletes' characteristics and other sources of data, and innovatively constructing a hybrid feature system that includes time series features and dynamic graph structure. Through the graph convolutional neural network modeling the time evolution characteristics of inter-country competition, combined with Bootstrap resampling technology to construct a probabilistic prediction model, to achieve the dynamic prediction of the medal list of the 2028 Los Angeles Olympic Games, and concluded that the United States with a probability of 85.7% to become the most progressive country, and Russia with a probability of 70.5% to become the most regressive country. This study establishes a multi-dimensional spatio-temporal correlation feature system and develops a hybrid GCN-Bootstrap architecture to realize the probabilistic inference of the Olympic medal distribution for the first time, and reveals the deep influence mechanism of geopolitical factors on competitive sports.