Optimal Recursive Reature Selection Method Based on Random Forest

To reduce the complexity of the model in feature selection and improve the classification accuracy to make feature selection more effective. In this paper, an optimal recursive feature selection method based on random forest is proposed. Multiple decision trees are constructed from random forest on the primed dataset, but the problem of overfitting occurs in the process of constructing decision trees, so search cross-validation is used to enhance the accuracy and credibility of the evaluation model. Meanwhile, the optimal recursive feature selection is obtained by the fuzzy information entropy method to eliminate noise. Cross-validation comparison is performed by dividing the training set 75% used for training and constructing the model to select the optimal parameters; the test set 25% used for testing the model to validate the assessment model. Finally, the validity of the model is verified by applying the example data of an undergraduate education index, and the feature indexes with importance greater than 0.01 are obtained by the importance calculation.