Abstract
Motivation is a key psychological factor influencing athletic performance, especially in high-intensity disciplines such as track and field. However, traditional assessment methods-ranging from self-report questionnaires to static physiological models-often fail to capture the temporal, individualized, and context-dependent nature of the motivation-performance relationship. In this study, we propose a hybrid EEG-based framework for modeling motivational states and forecasting athletic performance. The framework integrates neural indicators of arousal and stress with contextual and biomechanical variables using a dual-attention predictive architecture and a personalized adaptation mechanism. Rather than focusing on static prediction, the model dynamically adjusts to individual athletes' cognitive and physical states across training scenarios. Experimental validation on four public datasets, including two movement-oriented sets (MoBI and HASC), demonstrates consistent gains over strong baselines, with up to 3.5% improvement in accuracy and 7.6% improvement in early fatigue prediction. These findings suggest that the proposed system can support personalized monitoring and adaptive training strategies in performance-driven environments.