Multi-Dimensional Prediction of Student Learning Performance and Feature Importance Analysis Based on XGBoost Algorithm

In modern education, accurately predicting students' learning performance and identifying influencing factors is an important challenge. Current methods often face problems such as high data dimensions and complex variables. This paper studies multi-dimensional prediction and feature importance analysis of student learning performance based on the XGBoost algorithm, aiming to improve prediction accuracy and reveal key influencing factors. First, students' multi-dimensional data are collected, including attendance, homework scores, class participation, etc; then, through data preprocessing, missing values and outliers in the data are cleaned and standardized. Next, the XGBoost algorithm is used to train the model, set the initial parameters, and continuously optimize the hyperparameters, including learning rate, maximum depth, and subsampling rate, through grid search and cross-validation. Finally, the model is used to analyze feature importance and identify key factors that affect learning performance through feature weights. The results show that the model's prediction accuracy reaches <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 2 \%}$</tex>. In the feature importance analysis, homework grades and class participation are identified as the most important features. The conclusion shows that XGBoost has significant advantages in predicting student learning performance and provides educators with an effective decision support tool.