AdapARF: an adaptive hybrid framework for olympic medal forecasting

Predicting Olympic medal counts is a complex task involving historical trends and multidimensional influencing factors. This paper proposes a novel hybrid prediction framework—Adaptive ARIMA-Random Forest (AdapARF)—which combines the temporal modeling strength of ARIMA with the nonlinear learning capabilities of Random Forests. In this model, ARIMA captures long-term dependencies in medal data, while Random Forest incorporates exogenous variables such as host nation advantage, participation rate, and prior performance. An adaptive weight adjustment mechanism is introduced to dynamically optimize the contribution of each sub-model, enhancing the framework’s adaptability in evolving data contexts. The model is trained on Olympic data from 1960 to 2020 and validated using 2024 outcomes. Experimental results show that AdapARF outperforms standalone models in both predictive accuracy and robustness. In addition to forecasting medal counts for the 2028 Olympics, the model can identify countries likely to improve or decline. These methods offer strategic support for national teams and host nations.