A spatio-temporal GCN–Transformer framework for team performance analysis in sports psychodynamics with correlation consistency and stability modeling

Sports analytics increasingly demands a comprehensive understanding of both individual athlete behavior and collective team dynamics. However, most existing approaches primarily focus on temporal patterns while overlooking the critical spatial interdependencies among athletes. To address this limitation, this study proposes a hybrid Graph Convolutional Network (GCN)–Transformer framework that jointly models spatial and temporal dynamics in team performance analysis. The proposed model integrates GCNs to capture inter-athlete relationships through dynamic graph construction, while the Transformer architecture learns long-range temporal dependencies from multimodal sensor data, including physiological, biomechanical, and motion-based signals collected during training sessions. The proposed framework provides an effective computational approach for modeling sensor-derived proxy indicators of team coordination, stress, and fatigue using multimodal sports sensor data. The framework is evaluated using standard classification metrics, including accuracy, precision, recall, and F1-score, alongside domain-specific psychodynamic measures such as Inter-Athlete Correlation Consistency (IACC) and Temporal Stability Score (TSS). Experimental results demonstrate that the proposed model achieves 94.19% accuracy and 93.18% F1-score, outperforming baseline models. Furthermore, the model has an IACC of 0.847 and a TSS of 0.213, which are complementary measures of coordination consistency and temporal variation of the predicted team-level states. Ablation studies confirm the effectiveness of integrating graph-based spatial learning with attention-driven temporal modeling. The explainability analysis improves the transparency of model predictions by identifying influential physiological and biomechanical features associated with the learned representations.