Hybrid transformer-based security risk classification for sports events using crowd flow and congestion analytics

Sports events are often held at large venues with complex movement patterns and different levels of congestion, making the job of security management and public safety difficult. Correct identification of security threats is significant to effective security prevention and good crowd control. This research proposes a fusion of transformer models for the classification of security risk in sports events based on indicators of crowd flow and crowd density in the stadium. The Sports Event Security Risk Classification dataset covers attendance, crowd density, congestion, security resources, emergency accessibility and infrastructure. In addition, several features have been derived from existing features to the system for security reasons to improve the predictive capability of the system. FT-Transformer based feature interaction learning combines with TabNet based feature selection, named TabFT-Net. The experimental results show that TabFT-Net surpasses all baseline models and obtains an accuracy of 99.53%, and an F1 score of 99.53%. Moreover, by using SHAP and LIME analysis, the interpretability of the model enhanced by enabling the identification of the most influential security indicators. The proposed framework is further validated under ten-fold cross-validation, extensive ablation studies and computational complexity analysis for robustness and practical efficiency.