An Empirical Investigation into Enhancing Competitive Advantages in Cheerleading Competitions through Explainable Artificial Int

his study aims to enhance the accuracy and transparency of technical evaluations in cheerleading competitions by leveraging explainable artificial intelligence (XAI) techniques, driving innovation and optimization in traditional judging systems. To address the characteristics of cheerleading competitions, the ResNeXt network was employed for action recognition and feature extraction from competition videos. ResNeXt, with its efficient multi-branch architecture, captures intricate differences between various movements, making it highly effective for analyzing complex motion data. Additionally, the Vision Transformer (ViT) model was integrated to leverage its robust global information processing capability, significantly improving the evaluation accuracy of team coordination and overall performance. To enhance the model\u2019s transparency, the study incorporated the Shapley Additive Explanations (SHAP) algorithm, a key XAI technique, to quantify and visualize the basis for scoring each technical action, ensuring fairness and interpretability in the evaluation process. The empirical study utilized video data from 500 cheerleading competitions for training and testing, evaluating the model\u2019s scoring accuracy and explainability. Results demonstrate that the intelligent evaluation model based on ResNeXt and ViT improves scoring accuracy by 8.8% compared to traditional methods. The model excels particularly in recognizing movement details and assessing team coordination. Using SHAP explanations, judges can intuitively understand the key factors influencing the model\u2019s scores, enhancing trust and transparency in the scoring process. Moreover, the model's real-time scoring capability provides valuable data support for athlete training and technical improvement, enabling more precise personalized training plans.