Research On Sports Risk Prediction Model Based On Deep Learning And Ensemble Learning

With the vigorous promotion of the state and government,sports have gradually become popular and daily,and become an indispensable part of people’s lives.However,due to the lack of basic knowledge about the mechanism,method,approach,process and intensity of sports,sports-related risk events such as injury,disease and even sudden death often occur,which seriously affect people’s physical and mental health and even threaten people’s life safety.In view of this,the thesis researches the sports risk prediction model based on deep learning,based on which a sports risk prediction system is designed and implemented.The main research contents are as follows:First,traditional statistical methods ignore the interaction effects between multiple predictors in order to ensure the stability and repeatability of the model,and cannot fully exploit the data.Meanwhile,the classical machine learning algorithms have poor prediction accuracy and insufficient processing and prediction ability for high-dimensional and large sample size data,which may lead to high model specificity.To address such problems,a sports risk prediction model based on automatic coding and convolutional neural network is proposed.The model fully considers various risk factors of different exercisers,applies BSL sampling technique for the equalization operation of the data set,and uses information gain to calculate the contribution of each risk factor to the risk category to screen high-quality indicators.Based on this,the AE-CNN model is constructed to classify and predict sports risks.By using automatic coding for feature extraction of motion risk factors,the feature components that can highly characterize risk are obtained.Then,a convolutional neural network with double convolutional layers and double pooling layer topology is constructed.The automatically encoded feature components are then combined with the topology of the convolutional neural network to output the motion risk prediction results.Compared with other algorithms,the proposed method is able to analyze and extract risk features effectively and has a higher prediction accuracy.Second,the prediction capability of a single model is limited,and individual models tend to have incomplete consideration of other factors,which can cause problems such as unstable prediction results or poor performance.At the same time,using manual methods to extract features will ignore some ways of constructing features,and there are difficulties in extracting combined features that have not appeared in feature engineering.To address such problems,a motion risk probability prediction model that integrates DeepFM and LightGBM is proposed.The model is based on preprocessing the original data and feature selection,and training the DeepFM model and LightGBM model respectively,among which,the DeepFM model can extract the complex correlation of features from the original data to learn the high-order combined features,and can realize the cross-learning of low-and deep-level features.The model can effectively overcome the dependence of traditional gradient boosting tree algorithms on large-scale data,and significantly improve the computational efficiency and prediction accuracy of the model while reducing the computational complexity of the model.The model finally uses the Stacking algorithm to take the prediction results of these two models,as new features,and input them into the meta-learner for training,and the meta-learner outputs the final prediction values.The experimental results show that the fusion model can effectively extract the feature importance and predict the risk probability for the sports risk factors,and the prediction is better than that of the single model.Finally,in order to verify the rationality,scientificity and usability of the sports risk prediction model,the sports risk prediction system was designed and implemented by analyzing the sports demand of sportsmen,detailing the implementation process of each functional module,and showing the operation effect diagram of the prediction system.The system can assist exercisers in sports risk prediction,and identify risk sources in time by analyzing risk factors,so that corresponding interventions can be implemented to ensure the successful achievement of sports goals.The exercise risk prediction model studied in the thesis and the designed exercise risk prediction system provide new ideas for personalized exercise program generation.