Feature selection method based on Filter selection algorithm and Wrapper selection algorithm

Abstract

The invention discloses a feature selection method based on a Filter selection algorithm and a Wrapper selection algorithm. The method comprises the steps that all feature subsets are imported, a variance method is adopted to screen divergent features, a Pearson correlation coefficient method is adopted to screen non-redundant features, a feature space search method is adopted to generate new feature subsets, a neural network training learning model is adopted to construct an evaluation standard of the feature subsets, and the feature subsets are output. According to the method, the advantagesof the Filter selection algorithm and the advantages of the Wrapper selection algorithm are combined, and by use of the complementary characteristics of the two algorithms, calculation cost is reduced while algorithm efficiency is improved.

Description

technical field [0001] The invention belongs to the technical field of machine learning, and in particular relates to a feature selection method based on a Filter and Wrapper selection algorithm. Background technique [0002] Feature selection is one of the important research topics in the fields of machine learning, pattern recognition and statistics. Feature selection refers to selecting the feature set that obtains the best performance of the corresponding model and algorithm. Usually, when we construct a machine learning algorithm, we can collect data information in many dimensions, but when the feature dimension reaches a certain level, putting all the features into the algorithm will bring a disaster of dimensionality. Difficult to achieve convergence, and may even overflow the calculation. When faced with this kind of problem, feature selection becomes very important. [0003] Generally speaking, data and features determine the upper limit of machine learning, and ...

AI Technical Summary

Problems solved by technology

[0009] 1. The method of Embedded mode depends on the machine learning algorithm, and its applicability is not high

[0010] 2. The Wrapper selection algorithm is inefficient because it needs to use each candidate feature subset to train the model for evaluation

[0011] 3. The filter selection algorithm is more efficient, but it is difficult to completely remove redundant features, and it is easy to remove some internal dependency features as redundant features, that is, some features have low distinguishing performance when they are processed individually, but as a whole has a strong ability to distinguish

Images