Research on evaluation algorithm of student physical training effect based on machine learning

The aim of this paper is to develop an effective method to evaluate the progress of students after specific physical training in order to provide data support for physical education in schools. Firstly, a large number of students' physical training data were collected from a number of schools, including physical fitness indicators such as height, weight standing long jump distance, 50-meter sprint performance and other dimensions, and combined with the implementation of training plans and other information, a comprehensive data set was constructed. During the stage of data preprocessing, statistical methods are employed to detect and manage missing values and outliers, thereby guaranteeing the quality of the data. Through feature engineering, a number of key variables are extracted to help predict, such as the physical difference before and after training, training frequency, and the length of each training. Next, we applied three different machine learning models: Among the trio of networks—the Long Short-Term Memory Network (LSTM), Generative Adversarial Network (GAN), and Deep Neural Network (DNN)—the LSTM is utilized to ascertain the optimal decision boundary for differentiating students of varying physical capabilities. By incorporating multiple decision trees, the GAN enhances the stability and precision of predictions. The DNN, on the other hand, employs multi-layer perceptron structures to discern intricate patterns within the data. Subsequently, a novel neural network model, designated as LGNN, is formulated through the amalgamation of these three models for effect assessment, and its superiority is validated on a genuine real-world dataset.