PREDICTION OF 2028 LOS ANGELES OLYMPIC MEDAL TABLE BASED ON MULTI-MODEL INTEGRATION
Keywords:
Olympic medal prediction, LSTM, XGBoost, Host-country effect, Feature importanceAbstract
This study addresses the complex challenge of predicting medal standings for the 2028 Los Angeles Olympics by developing an innovative hybrid modeling framework that integrates time-series analysis, machine learning regression, and Bayesian probabilistic approaches. Building upon previous Olympic prediction research, we propose a multi-layered architecture that combines the temporal processing capabilities of Long Short-Term Memory (LSTM) networks with the feature importance quantification of XGBoost/LightGBM algorithms and the uncertainty modeling of Bayesian hierarchical frameworks. The research incorporates an unprecedented range of predictive features, including historical medal performance (1996-2024), athlete participation metrics, event-scale characteristics, and quantified host-country advantages, to construct a comprehensive predictive system. Our empirical results demonstrate that the United States is projected to maintain its dominance with 40 gold medals (128 total), benefiting significantly from host-country effects (SHAP value +12) and established strengths in swimming and track and field (contributing 43% of gold medal variance). China shows steady growth to 35 gold medals (95 total), while the United Kingdom and Japan exhibit strategic gains in cycling and skateboarding respectively. The model achieves superior predictive accuracy (R²=0. 89 for gold medals) compared to traditional ARIMA approaches, with sensitivity analysis revealing three key insights: (1) track and field and swimming remain the highest-yield events for medal acquisition, (2) host nations experience a quantifiable 15-20% medal boost through venue familiarity and optimized scheduling, and (3) emerging economies demonstrate diminishing marginal returns on GDP investments beyond $5 trillion. The framework provides actionable intelligence for Olympic stakeholders, enabling data-driven resource allocation between sports disciplines and offering probabilistic projections for underdog nations (e.g., Malaysia with 80% probability of first medal in badminton). Methodological innovations include event-level feature engineering, game-theoretic modeling of coaching migrations, and policy-specific recommendations tailored to nations at different development stages.References
[1] Han Z, Zhao J, Leung H, et al. A review of deep learning models for time series prediction. IEEE Sensors Journal, 2019, 21(6): 7833-7848.
[2] Fahrmeir L, Kneib T, Lang S, et al. Regression models//Regression: Models, methods and applications. Berlin, Heidelberg: Springer Berlin Heidelberg, 2022: 23-84.
[3] Cai W, Wen X, Li C, et al. Predicting the energy consumption in buildings using the optimized support vector regression model. Energy, 2023, 273: 127188.
[4] Veenman M, Stefan A M, Haaf J M. Bayesian hierarchical modeling: An introduction and reassessment. Behavior Research Methods, 2024, 56(5): 4600-4631.
[5] Spurio Mancini A, Docherty M M, Price M A, et al. Bayesian model comparison for simulation-based inference. RAS Techniques and Instruments, 2023, 2(1): 710-722.
[6] Knuepling L, Broekel T. Does relatedness drive the diversification of countries’ success in sports?. European Sport Management Quarterly, 2022, 22(2): 182-204.
[7] Hilbe J M. Logistic Regression//International Encyclopedia of Statistical Science. Berlin, Heidelberg: Springer Berlin Heidelberg, 2025: 1386-1390.
[8] Yang Y Y, Rashtchian C, Zhang H, et al. A closer look at accuracy vs. robustness. Advances in neural information processing systems, 2020, 33: 8588-8601.
[9] Mulder J, Raftery A E. BIC extensions for order-constrained model selection. Sociological Methods & Research, 2022, 51(2): 471-498.
[10] van Putten I, Dichmont C M, Dowling N A, et al. Interconnected partnerships: Mapping collaborations in Australian fisheries stock assessment. Fisheries Research, 2025, 282: 107281.