A Real-Time Athlete Score Prediction Using Convolutional Neural Networks to Improve Accuracy Compared to Recurrent Neural Networks

The primary objective is to develop a Convolution Neural Network-based real-time athlete score prediction system that outperforms RNNs in terms of accuracy. The work attempts to provide more precise and timely forecasts, which are crucial for efficient sports analytics, by utilizing the spatial nuances seen in athlete score photographs. We compiled a diverse collection of annotated athlete score photos from several sports settings in an effort to achieve real-time athlete score prediction. We designed an optimized Convolutional Neural Network (CNN) architecture including convolutional layers, pooling layers, and fully connected layers, and adapted it for the extraction of spatial features. The Convolution Neural Network was carefully trained with the Adam optimisation method, and its accuracy and loss metrics were continuously monitored. A comprehensive comparison study utilizing Recurrent Neural Networks (RNNs) evaluated important performance metrics such as precision, computational effectiveness, and instantaneous inference capabilities. Convolution Neural Network is a powerful tool for improving real-time sports analytics. This was demonstrated by the integration of the trained Convolution Neural Network model into a real-time athlete scoring system, which involved optimizing processing pipelines for low-latency performance without sacrificing prediction accuracy. The Convolution Neural Network based feature extraction technique has better accuracy of 89.28% than the Recurrent Neural Network with 79.41 %. The statistical significance value obtained is 0.760 (p>0.05),hence it is observed that there is no statistical significant difference between the two groups. The sample records were measured with the help of clinical analysis, the value of alpha is 0.05, the value of beta is 0.5, percentage of confidence is 95. Convolutional Neural Network algorithms provided better results in Athlete Score Prediction than Recurrent Neural Network algorithms.