OLYMPIC MEDAL PREDICTION BASED ON TPE-SEQ2SEQ MODEL

This paper proposes an innovative TPE-Seq2Seq model for Olympic medal prediction by integrating sequence-to-sequence deep learning with Tree-structured Parzen Estimator hyperparameter optimization. Utilizing historical Olympic data from the International Olympic Committee, we first constructed a comprehensive dataset through BP Neural Network-based imputation of missing values and integration of non-medal-winning nations. The model captures complex temporal patterns and feature relationships through an encoder-decoder architecture, with key hyperparameters (learning rate, hidden units, regularization coefficients) systematically optimized via TPE to mitigate overfitting and enhance generalization. Experimental results demonstrate significant performance improvements, achieving an R2 value of 0.875 on the test set. Monte Carlo simulation and 95% confidence intervals quantify prediction uncertainty, revealing stable forecasts for six leading nations at the 2028 Los Angeles Olympics. Notably, the model predicts 41 gold medals for the United States and 40 for China, with narrow confidence intervals (e.g., US gold: [39,42]), demonstrating high reliability. This data-driven framework offers strategic insights for national Olympic committees and event organizers in resource allocation and competition planning.