Software package for predicting swimming competition results based on language models of deep learning neural networks

Relevance. For an athlete, in addition to effective training, it is important to assess his functional state and readiness to show the best result when performing at competitions. At the same time, the assessment of the athlete's condition should be fast, low-cost and without distraction from the competitive process. Goal. Development of an innovative software package using deep learning methods and NLP (Natural Language Processing) to accurately predict the athletic performance of swimmers. The package is designed to support effective decisions on the selection of the strongest athletes for participation in team and individual swims based on the analysis of their functional states through automated conscious dialogue and taking into account key hematological indicators. Methods. Real data on athletes of the Russian national swimming team were used to train the system - a total of 100 cases were analyzed for eight key hematological parameters affecting athletic performance. To create the package, deep machine learning neural networks were used based on the Keras open source library and the Python language. Results: The developed complex includes three main modules: module "Training": creating a language model based on the Keras and Python libraries; module "Chat + Forecast": supporting a meaningful dialogue with the user (trainer / doctor / athlete) regarding the current health status and level of readiness; module "Test-Forecast": mass testing of the effectiveness of the developed model on specific examples. The accuracy of the model reached 90% during training based on the analysis of the model training history. Real testing confirmed the effectiveness of the approach, showing the coincidence of the forecast with the results in 87% of cases. Conclusion. The developed complex provides a quick assessment of the athlete's condition, minimizing costs and allowing for the effective selection of participants for various competitions. Possibilities for increasing accuracy are preserved by expanding the data set. A chatbot with natural language queries is used to obtain various data on the subject area.