Feature selection embedded cluster distribution position for characteristic analysis of multi-dimension poverty-stricken households in China

• Published in: Journal of Applied Science and Engineering
• Main Author: Liu H.; Liu Y.; Zhang R.; Liu D.; Zhang Z.
• Format: Article
• Language: English
• Published: Tamkang University 2021
• Online Access: https://www.scopus.com/inward/record.uri?eid=2-s2.0-85104272052&doi=10.6180%2fjase.202106_24%283%29.0003&partnerID=40&md5=e502b3f72ac48700641a039b9c5346be

Poverty is a historical problem all over the world. Poverty alleviation targeted to the primary problems faced by the poverty-stricken households in different classes can effectively improve the efficiency of poverty reduction. Owing to the large number of features that reflect the poor situation of poverty-stricken households, it is difficult for traditional methods to accurately analyse the primary problems faced by the poverty-stricken households. In view of the high performance of feature selection methods in dealing with high-dimensional data, a feature selection approach by taking the distribution position of features in each class into consideration is proposed in this paper. We use the Gaussian mixture model to describe the distribution of features in the same dimension, and measure the distribution position of the cluster consists of features in each class according to their Gauss mixture ingredients. Features with significant differences in distribution position between classes are selected, which can effectively represent the characteristics of samples in different classes. According to the experimental results, the proposed method performs well in determining the characteristics of samples in different classes, and can accurately analyze the typical features of poverty-stricken households in different classes, which provides the basis for the design of targeted poverty alleviation strategies. © 2021 Kalmyk Scientific Centre of Russian Academy of Sciences. All right reserved.