Research on gold medal prediction model based on multi-algorithm fusion

Abstract

In this study, a prediction framework based on multi-source historical data is proposed, and feature engineering and regularization techniques are used to construct interpretable machine learning models. After screening key features through Pearson correlation analysis, ridge regression and SHAP prediction models are built respectively, and the main effect correction parameter is innovatively introduced to enhance the model generalization ability. In the process of feature space construction, dynamic weight indicators are generated based on time-series data, and high-dimensional feature dimensionality reduction is accomplished through logistic regression, ultimately forming a predictor system containing four core dimensions. At the algorithmic level, this study compares the performance of L2 regularized ridge regression and game theory-based SHAP prediction model in prediction. And the experiments are parameter tuned, and the results show that the regularized model is effective in suppressing overfitting, and its mean square error (0.281) is reduced by 71.4% compared with the SHAP benchmark model. By constructing a dual assessment system, both the prediction accuracy (R²>0.95) and the quantitative resolution of feature contribution are ensured. In terms of feature importance analysis, the SHAP value calculation reveals the nonlinear relationship of each dimension, and the probability prediction module realizes the interval estimation of the probability of event occurrence through logistic regression integration. The framework exhibits strong robustness on the validation set. The hybrid modeling approach proposed in this study provides a new technical path for the time-series prediction problem, and its modular design can be extended to other prediction scenarios that require a balance between accuracy and interpretability.