An explainable dual-attention transformer for predicting the sociocultural impact of global sports events

Existing research on sociocultural analytics in global sports has primarily relied on traditional statistical and regression-based approaches that capture limited linear relationships among cultural, economic, and sentiment-related variables. However, the societal influence of international sporting events emerges from complex, nonlinear interactions that require more expressive modeling frameworks. This study proposes a Sociocultural Dual-Attention TabTransformer (SC-DATransformer) to predict a continuous Social Impact Score, defined as a normalized composite index derived from validated indicators including cultural engagement, diversity, gender representation, public sentiment, sustainability, and media coverage. The empirical analysis is conducted on a structured dataset comprising approximately 70,000 event-level observations described by 21 variables, covering multiple global competitions across regions and years. To enhance representation quality and reduce redundancy, feature engineering and selection techniques such as mutual information and dimensionality reduction are employed. Model performance is evaluated against established baselines using complementary regression metrics, supported by 95% bootstrap confidence intervals and Diebold–Mariano statistical testing. Within the experimental setting, the proposed model demonstrates improved predictive consistency and lower error magnitudes relative to baseline approaches. Explainable AI analysis further provides transparent insight into dominant sociocultural drivers influencing impact estimation.