A covariate aware and dual convolutional network model for stadium crowd flow prediction.

Area of Science:

• Sports Science
• Data Science
• Educational Technology

Background:

• Campus sports facility utilization shows significant temporal fluctuations impacting resource allocation and student exercise.
• Overcrowding during peak hours reduces teaching quality and increases injury risk, affecting participation sustainability.
• Accurate prediction of student participation is crucial for optimizing physical education schedules and enhancing efficiency.

Purpose of the Study:

• To develop a crowd flow prediction model specifically for physical education scenarios.
• To improve the accuracy and robustness of predicting student participation trends in sports facilities.
• To support data-driven optimization of teaching resources and physical education management.

Main Methods:

• Proposed a novel crowd flow prediction model integrating a covariate-aware cross-attention mechanism and a dual-layer convolutional feedforward network (DConvFFN).
• Incorporated external covariates like time, temperature, and academic schedules to capture complex relationships with participant flow.
• Employed cross-attention to extract interactive information between temporal features and external variables, and DConvFFN for enhanced feature learning.

Main Results:

• The model effectively identified complex periodic patterns and sudden fluctuations in crowd trends within physical education settings.
• Achieved a 6.16% reduction in Mean Squared Error (MSE) and a 4.06% reduction in Mean Absolute Error (MAE) compared to traditional RNN and Transformer models.
• Demonstrated superior accuracy and robustness in predicting crowd flow in sports venues.

Conclusions:

• The proposed model offers significant improvements for predicting crowd flow in physical education environments.
• Results highlight the model's potential for practical applications in pre-allocating teaching resources and dynamically adjusting curricula.
• Provides key technical support for building data-driven smart sports teaching management systems.