Binary Equalization Variable Order Trans High Dimensional Fea Optimizer Based on fer Functions to Solve ture Selection Problem

Feature selection (FS) is the core concept in the field of both machine learning and data management. FS can eliminate irrelevant or partially related features to improve model performance. Wrapped FS method typically solves the discrete optimization problems by converting continuous values to binary values based on a transfer function. A binary equilibrium optimizer based on variable-order transfer function was proposed to solve high dimensional feature selection problem. Based on the four basic transfer functions of V-shaped, I -shaped. S-shaped and /-shaped, eight transfer functions with varying order are proposed. Based on the binary equilibrium optimizer, the Relief guided strategy and the KNN classifier, the wrapped feature selection was realized. The simulation experiments were divided into two groups by using 12 medium and high dimensional standard UCI datasets. The first group analyzes the feature selection effect of 8 transfer functions under variable order parameters. In the second group of experiments, each winning parameter in the first group of experiments is selected for comparison. The experiment results prove that a reasonable setting of the order of transfer function can obtain better optimization results, and By analyzing the laboratory data, it can be concluded that the feature selection performance under Sl(0.4), S2(l), Zl(2) and V2(3) is better.