Feature extraction and feature selection methods for background multi-source data

Abstract

A feature extraction and feature selection method for background multi-source data is characterized in that it includes the following steps: (1) dividing training sets and test sets on the background data of multiple months; (2) training sets for different Extract the corresponding group features from the source data; (3) Use the Group Lasso method to select feature groups through cross-validation on the test set. The beneficial effects of the present invention are as follows: for the selected group features, the C45 decision tree is used to establish a classifier for off-network user analysis classifier, and the accuracy rate of off-network user prediction reaches 45%. The accuracy rate reached 88%.

Description

technical field [0001] The invention relates to a background multi-source data-oriented feature extraction and feature selection method for off-network user analysis. Background technique [0002] For each household's daily online time series, there is currently no good way to characterize the changing trend characteristics of users' online time. The Lasso method is a sparse feature selection method. When Lasso is directly applied to a model with a group structure, it tends to select a single feature and destroys the group structure of the feature. For LR classifiers, small changes in features will also have a great impact on the final prediction results. [0003] The Group Lasso method introduces the extension of the penalty function to study the selection of group features. The Filter method is a feature selection method that has nothing to do with the learning machine, and selects a subset of features through a certain measure. A commonly used measure is the Pearson co...

AI Technical Summary

Problems solved by technology

For LR classifiers, small changes in features will also have a great impact on the final prediction results

Images