Partially-weighted incomplete data hybrid clustering method

Abstract

The invention relates to a partially-weighted incomplete data hybrid clustering method, and the method comprises the following steps: (1), collecting data; (2), carrying out data processing: randomlyenabling a part of data to be short of a part of attributes, and enabling the part of data to become incomplete data; (3), carrying out the data estimation: carrying out the estimation of the lost data through an individual in an improved genetic algorithm; (4), carrying out the clustering analysis: carrying out the fuzzy clustering analysis of the estimated data. The invention proposes a partially-weighted incomplete data hybrid clustering method (GLW-FCM) optimized through the improved genetic algorithm, and achieves a purpose of searching all problem spaces to find an optimal solution. A UCI standard test dataset comprises Iris, Bupa, Wine, and Breast. The contrast experiment of the method provided by the invention with other five algorithms in a Matlab environment is carried out, and the improved algorithm hides the parallelism in the whole problem space for the searching of the optimal solution, and obtains the more ideal clustering result. The method effectively reduces the meanerror classification number, the mean error classification standard deviation and the mean iteration ending number.

Description

technical field [0001] The invention relates to a locally weighted hybrid clustering method for incomplete data, which belongs to the field of incomplete data clustering. Background technique [0002] The rapid development of information technology has brought massive and complex data in various fields, which has far exceeded the ability of human beings to process both in terms of capacity and scale. In order to efficiently and accurately analyze these data, cluster analysis has become a new development and new trend, using computer programs to intelligently and accurately complete the classification. [0003] Fuzzy C-means (FCM), as a basic unsupervised clustering method, is usually suitable for clustering data without missing attributes. However, in real life and industrial applications, factors such as data leakage, input errors, equipment failures, plan changes, data acquisition failures, and random noise effects cause the data to be incomplete. At this time, the data w...

AI Technical Summary

Problems solved by technology

[0004] If you only use local weighting to optimize the clustering algorithm, similar to the FCM algorithm, it uses a gradient descent mechanism to optimize; it is more sensitive to the initial parameter values ​​​​and is prone to local convergence limitations, etc.

If only the genetic algorithm is used to optimize the clustering algorithm, although the clustering effect is significantly improved, there are still defects such as premature convergence.

Images