Research And Application Of Attribute Reduction Algorithm In The Incomplete Mixed Decision System

In order to deal with large amounts of data effectively and obtain potentially valuable knowledge, data mining has become a hot topic of research. Data classification is an important data analysis technique in the domain of data mining, it can be used to extract models of describing important data classes or predict future data trends. Rough sets, as a kind of mathematical tool of handling fuzzy and uncertainty, have become an important method for data classification.There being such a decision system with missing attribute values or the mixed data types or both of them, the classical rough sets theory cannot directly do anything about it, because the classical rough sets theory have been proposed based on the equivalence relation. This article proposed a corresponding solution to this problem.The main innovation of this article:1) On the basis of the classical rough sets theory, with neighborhood rough sets as the starting point, combined with extended rough sets model in the incomplete decision systems, the limited neighborhood relation and the extension neighborhood relation was proposed, the nominal attribute and the numerical attribute and the missing attribute could be handled simultaneously by the proposed two kinds of relation model without the discretization of numerical attributes or completing the incomplete data, thus it eliminated the uncertainty of data preprocessing for these types of data.2) Based on the limited neighborhood relation and the extension neighborhood relation, according to the idea of algebraic theory, information theory and discernibility matrix, to design attribute reduction algorithm correspondingly. The proposed reduction algorithm was tested on several UCI data sets. The experiment results show that the rationality of the limited neighborhood relation and the extension neighborhood relation and the effectiveness of the reduction algorithm. Also it was discussed that how to impact the classification when specifying the values of the threshold parameters used in the limited neighborhood relation and the extension neighborhood relation.3) According to the problem of movie box office forecasting lack of scientific theories guiding in China, the two kinds of new relation model and four corresponding attribute reduction algorithm proposed above, were applied to box office prediction model. The research could eventually provide relevant value judgments and decision support for investors or the producers of the film, and also provide a reference to predict future trends in other areas or markets.