Medal Prediction Models Based on LASSO Regression and Random Forest Algorithm

Abstract

Medal prediction serves as a critical research direction in sports science and data analysis, holding significant implications for optimizing resource allocation and strategic decision-making in competitive sports. This study proposes an innovative hybrid predictive model that integrates hierarchical clustering, LASSO regression, and random forest algorithms. By constructing a purely competition-endogenous multidimensional competitiveness indicator system, the model overcomes the limitations of conventional approaches that rely heavily on external factors. The methodology begins with establishing feature-based indicators to categorize participating nations into three distinct clusters through hierarchical clustering, reflecting their respective stages of sports development and establishing an optimized differentiated modeling framework. For countries at different developmental stages, LASSO regression and random forest algorithms are strategically applied, achieving both model robustness and systematic exploration of feature importance. Empirical results demonstrate the model's capability to accurately forecast medal distributions for the 2028 Los Angeles event, with predictions aligning closely with historical trends and prediction errors confined within a margin of 2 medals. This research provides a quantifiable decision-making tool that substantially enhances the scientific basis for event resource allocation and policy formulation in competitive sports systems.