CNN and RF based high-dimensional multi-granularity feature selection method

Abstract

The invention relates to a CNN and RF based high-dimensional multi-granularity feature selection method, and belongs to the technical field of information processing. The method is based on the high-latitude multi-granularity feature data set, and is combined with the deep learning algorithm and the machine learning algorithm to solve the problem of high-latitude multi-granularity feature extraction. Firstly, a deep learning algorithm CNN model is used for constructing an FSelCNN model, original data are converted into single granularity from multi-granularity through the model, and the data are made to become data needed by a machine learning algorithm; and finally, effective features influencing the actual problem are selected from the high-latitude data by using a machine learning algorithm RF. Starting from the single feature level of the high-latitude multi-granularity feature data, the multi-granularity dimension of the high-latitude multi-granularity feature data is converted into the single-granularity dimension, so that the operation complexity is effectively solved; the model reduces the parameter quantity and can complete training in a short time; and the method is suitable for various high-weft multi-granularity data, has relatively high adaptive capacity and has a relatively good effect.

Description

technical field [0001] The invention belongs to the technical field of information processing, and relates to a high-dimensional multi-granularity feature selection method based on CNN and RF. Background technique [0002] With the explosive growth of data in the Internet era, various forms of data characteristics have emerged, and an efficient method is urgently needed to solve the problems caused by various forms of data, so as to better provide efficient data support for machine learning models , and effectively reflect the actual effect brought by the data. Moreover, feature engineering plays an irreplaceable role in the actual application of machine learning. In the field of machine learning, it is generally believed that the upper bound of machine learning algorithms depends on data and feature engineering, and the final model is just to continuously approach this upper bound through linear and nonlinear methods. [0003] Therefore, how to do a good job in feature en...

AI Technical Summary

Problems solved by technology

[0004] However, although the above algorithms solve the dimensionality disaster problem in terms of dimensions, they ultimately combine features through linear combinations of some features and extract the most influential features. They are not specific to a certain feature, which does not satisfy some Specific problems, such as extracting 25 parts from 42 parts of the human body, and balancing the body balance ability of the elderly through these 25 parts

For this type of problem, traditional PCA and LDA methods can no longer meet its needs

Images