Action Assurance: Federated CNN Architectures Decoding Cricket Umpire Signals
Ankita Suryavanshi, Shiva Mehta, Preeti Madhukar Chaudhary, Siddhant Thapliyal · 2024
This article shows a federated learning framework based on CNN and will identify cricket umpire gestures. We have federated and used Federated training mechanics to train the five clients: WACA, BBL, CPL, IPL, and SMA. This way, we applied a wide range of data sources with an equal platform to maintain the anonymity of each person's data. The model accurately categorized five unique umpire signals: ‘Out,’ ‘Four,’ ‘Six,’ ‘No Ball,’ and ‘Wide Ball.’ This assessment consisted of four measures, e.g., precision, recall, Fl-score, and accuracy. The performance of the five customers' C-Ievel outcomes was between 75.63% and 87.05% regarding the overall average accuracy. Among these, datasef _ 5 is the most suitable for model training, showing the highest scores in rating all the metrics. The result is the accuracy (the weighted average), which changes between 75.64% and 87.07% Precision. The precision computed using the micro average ranged from 75.65% to 87.02 %, and it was also found that 30.36% of sentences gained improvements in terms of pyrotechnical relevance. The results confirm the model's reliability. An overall accuracy of 92% shows a high level of accurate world application adaptation. The accuracy of the ‘Out’ signal has shown the highest precision. Since this signal brings forward consequent outputs, successful implementation of the model will be associated with the highest efficiency in payouts from the system. The federated learning method was introduced to solve this issue, using a technique that makes it impossible to reconstruct client data. Individual client data is from the changes.