Research on optimization and system application of improved CART algorithm for multidimensional enrollment data

CART algorithm, as a classical decision tree algorithm, faces challenges such as unbalanced categories and insufficient feature selection when dealing with multidimensional enrollment data. In this paper, an improved CART algorithm for multi-dimensional enrollment data is proposed. The features are jointly screened by MI-L1 regularization, the split criterion is optimized by introducing dynamic weighted Gini index, the dynamic threshold pruning strategy driven by cross-validation is adopted, and the square error method weighted by feature importance is adopted for regression tasks. The experiment is based on the enrollment data of a key university in a province. The results show that the improved CART algorithm significantly improves the classification accuracy and recall rate, improves the efficiency of feature selection, and the dynamic threshold pruning strategy performs best. The further designed multi-dimensional enrollment data processing system includes four modules: data preprocessing, feature selection, model construction and result display, which can effectively manage and analyze enrollment data, provide accurate and personalized student evaluation and suggestions for educational institutions, and promote the development of education.