Hits:

Date of Publication:2022-10-04

Journal:小型微型计算机系统

Issue:7

Page Number:1523-1527

ISSN No.:1000-1220

Abstract:Data imputation is an important issue in data processing and analysis which has serious impact on the results of data mining and learning.Most of the existing algorithms are either for the same type attributes or difficult to determine the parameters.Aiming at these problems,the paper proposes a multi-kernels function based algorithm to fill mixed attributes data.In order to reduce interference and computation,denoising deep belief network with rectified linear units model is developed for feature extraction from incomplete data and clustering.Specially,to reduce the times of iteration,partial-distance strategy is used for missing values initialization so that results can be rapid convergence and more accurate.After calculating the probability density function about missing variable and complete variable,an estimator which is easy to determine parameters is constructed for missing value prediction.Through the experiment results,the algorithm can reduce the complexity of determining parameters and iteration times,at the same time ensure accuracy.

Note:新增回溯数据