Deep learning method based on feature clustering

Abstract

The invention discloses a deep learning method based on feature clustering. The method comprises the following steps: selecting characteristic variables from the specific data set, carrying out data preprocessing on the selected characteristic variables, and calculating correlation coefficients among the characteristic variables; screening characteristic variables with high correlation coefficients by utilizing a custom function, screening out principal components in the characteristic variables, carrying out construction and clustering processing on the screened-out principal components, andguiding neural network construction based on a clustering result. According to the invention, data preprocessing is carried out on selected characteristic variables. Data scaling can eliminate differences of characteristic attributes such as characteristics and orders of magnitudes among different samples. Through dimension reduction processing, the samples can be mapped to a low-dimension space to be displayed. The optimal clustering mode can be conveniently selected according to the shape of the observation data in the later period, the accuracy of feature clustering can be improved, the feature variables with high correlation coefficients are screened through the user-defined function, and the problem that the correlation of the selected independent variables in clustering analysis is low can be solved.

Description

technical field [0001] The invention relates to the technical field of machine learning, in particular to a deep learning method based on feature clustering. Background technique [0002] In the field of deep learning, the mainstream deep learning architectures include DNN, RNN and CNN. DNN is a neural network with fully connected features, which is a general deep learning method; RNN is a recurrent neural network, which is also a fully connected structure. For scenarios with temporal context, such as the NLP field; CNN is a convolutional neural network, characterized by local connections based on spatial correlation, and is mainly used in the field of image processing. At present, the advantages and disadvantages of these three mainstream neural network structures are also very obvious. The feature local correlation connection of CNN reduces a lot of parameter storage and calculation; DNN does not consider feature correlation, and directly connects all features directly, ca...

AI Technical Summary

Problems solved by technology

[0004] In response to the above problems, the present invention proposes a deep learning method based on feature clustering. By performing data preprocessing on the selected feature variables, data scaling can eliminate the differences in characteristic attributes such as characteristics and magnitudes between different samples, and reduce dimensionality. Samples can be mapped to a low-dimensional space for display, which is convenient for later selection of the most suitable clustering method based on the shape of the observed data, and can improve the accuracy of feature clustering. By using custom functions to filter feature variables with high correlation coefficients, you can Solve the problem of relatively low correlation of selected independent variables in cluster analysis