Application of Big Data Technology in Sports Activity Participation Behavior Analysis System

Current sports data comes from various sources, with large data volume and high complexity, making it difficult to extract valuable information. This paper introduces the Convolutional Neural Network (CNN) model and the Long Short-Term Memory (LSTM) model to optimize the sports activity participation behavior analysis system. This paper extracts the features of sports videos and sensor data through convolutional neural network classification, combines K-Means clustering algorithm to group user sports behaviors, and provides personalized sports recommendations. LSTM is used to predict user sports behaviors and dynamically adjust sports plans. Through data preprocessing, fusion and feature extraction, the accuracy of motion state assessment is improved, and personalized motion planning and behavior prediction are optimized. Experimental results show that ResNet34 performs well in motion behavior recognition, with a classification accuracy of $90.1 \%$ and a test accuracy of $87.2 \%$, which is better than Basic CNN and Visual Geometry Group 16-layer (VGG16). The LSTM model has a prediction accuracy of $91.5 \%$ in the high-activity group, and the adjustment effect of exercise intensity is $8 \%$. The prediction accuracy of the low-activity group is $79 \%$, and the adjustment effect is only $3 \%$. The CNN model has strong classification performance, and LSTM can effectively predict and adjust the exercise behavior of different groups. This study provides technical support for personalized sports recommendations and accurate predictions, and promotes the increase in sports participation and the promotion of a healthy lifestyle.