Academic Performance Prediction Based on Hybrid GA-XGBoost-FM Learning

Considering the issues of data imbalance and sparsity in hybrid learning performance classification prediction, as well as the need to improve model performance, this study proposes a GA-XGBoost-FM hybrid learning performance classification prediction model. First, the SMOTE sampling method is used to balance the dataset. To address data sparsity, XGBoost is applied for feature crossing, while Genetic Algorithm (GA) is utilized for hyperparameter optimization during this process. The leaf nodes of the generated trees are then one-hot encoded to create high-order feature data, which is subsequently fed into the Factorization Machine (FM) for iterative training. Experimental results indicate that the GA-XGBoost-FM model achieves an accuracy of 90.59% in hybrid learning performance prediction. This model effectively classifies and predicts student learning outcomes, providing valuable insights for enhancing teaching effectiveness.