Research On Prediction Of Win Rate Of Ultimate Fighting Championship Based On Recurrent Neural Networks

The Ultimate Fighting Championship(UFC)is one of the world’s most popular professional mixed martial arts competitions,and the rapid growth of UFC competitions has generated tremendous commercial benefits,but also posed a higher challenge to UFC clubs and competition operators in terms of decision making,for which effective prediction of win rates is an excellent tool.Based on the fact that traditional prediction models have difficulty in fitting the nonlinear data generated in the game,this paper aims to improve the accuracy of UFC event win rate prediction by constructing a recurrent neural network prediction model based on grouping strategy and spatial feature fusion,and providing an iterative framework for UFC event win rate prediction based on the prediction results.The main research contents are as follows:(1)In this paper,UFC competition feature clusters are established by data mining and feature extraction methods.The original information is obtained from the UFCSTATS website for feature engineering,and the correlation weights between variables are obtained by analyzing the correlation of variables,and the semantic features of discrete data are mined by using Doc2 vec and one-hot coding techniques to better characterize the deeper features of data.In addition,this paper uses the XGBoost model to analyze the importance of different features and selects ten features for weighting to further improve the prediction accuracy of the model.(2)A novel grouping strategy is proposed in this paper.Recurrent neural networks feature that current features can be combined with previous feature information to make predictions,but when two groups of features are far away the connection is weak and less information is retained,which will not only reduce the prediction accuracy but also increase the computational complexity of the model.To address the above situation this paper proposes a grouping strategy that can enhance the contextual information between features with stronger relationships.(3)A spatial feature fusion strategy is proposed in this paper.The traditional feature fusion strategy usually splices the feature sequences directly,resulting in the inability to explore the spatial relationship between feature information.In this paper,a spatial feature fusion strategy is proposed to fuse each column of the feature map across spatial dimensions to fully extract deep features,which can effectively improve the accuracy of the prediction model.Based on the above work this paper proposes a UFC event win rate model based on grouping strategy and spatial feature fusion,which can effectively predict the win or loss of UFC matches by obtaining contextual semantic information over long distances of UFC temporal data.Based on this work,this paper designs an iterative framework for UFC event win rate prediction task,with UFC big data and prediction model as two central points,UFC event operators and clubs to improve the accumulation and expansion of UFC big data,and scholars and researchers to continuously improve and expand the prediction algorithm,so as to form a benign development of big data and prediction algorithm.The research results show that: the proposed grouping strategy and spatial feature fusion effectively improve the accuracy of tournament win rate prediction;(2)the proposed tournament win rate prediction algorithm based on grouping strategy and spatial feature fusion achieves outstanding results in both simulation and comparison experiments;(3)the method in this paper is a general technique that also has positive reference value for related industries.