Based on Multimodal Time Series Models: Olympic Medal Prediction and National Competitiveness Analysis

This study addresses the critical challenge of long-term trend and multi-source coupled modeling in Olympic medal prediction by proposing a Multimodal Fusion Frame-work (MDFF). The framework incorporates: (1) a 26-feature national competitiveness evaluation system optimized via Recursive Feature Elimination (42% redundancy reduction), (2) a hierarchical prediction mechanism combining LSTM-TimesFM for decade-scale trend capture and SMA-TTM hybrid model for quadrennial fluctuation optimization, supplemented by (3) a GBDT-LR cascade model resolving small-sample limitations. Innovatively, we develop an Isolation Forest-Granger causality testing method for coaching effect quantification, demonstrating elite coaches’ significant performance enhancement in specific disciplines (peak GCE index=0.892). Experimental results show 79.1% F1-score and AS=0.367 sensitivity index for 2028 Olympic predictions. This research provides international sports governance with a scalable quantitative analysis tool.