BP Neural Network Models for Multilevel Optimization: Combining Genetic Algorithms with Regularization Strategies for Performance Analysis

Analysis of student behavior is an essential study in research in the field of education. Among them, student behavior is often highly random and unpredictable. It is a challenge to achieve more accurate prediction of student behavior so as to regulate and guide student behavior in decision making and implementation. Therefore, this paper proposes a BP neural network model based on the combination of genetic algorithm and regularization technique to optimize the accuracy and robustness of behavior prediction in colleges and universities. The model effectively suppresses the overfitting of the network and improves the generalization ability of the model by introducing regularization means such as L2 regularization and Dropout. Meanwhile, the genetic algorithm is used to automate the adjustment of key hyperparameters in the BP neural network, including the regularization coefficient, the learning rate, and the number of neurons in the hidden layer, in order to obtain the best model performance. The significant advantages of the method in terms of prediction accuracy, model stability, and computational efficiency are verified by comparing with traditional neural networks and conventional regression models. The results show that the BP neural network model combining genetic algorithm and regularization technique achieves higher accuracy in the task of college grades prediction and demonstrates strong advantages in preventing overfitting and optimizing hyperparameters. In addition, this paper explores the strategy of multilevel optimization, demonstrates the effectiveness in high-dimensional data processing and complex model optimization, and provides valuable references for the construction of future educational data analysis and intelligent decision-making systems.