Research On The Rough Set Model Of Incomplete Decision Information System With Hybrid Value And Reduction Algorithm

With the rapid development of science and technology and the continuousimprovement of automation level, massive complex information systems have appeared in many areas. These information systems not only include discrete data but also include continuous data, referred to the information systems include mixed data. If we take advantage of the equivalence relation of the classical rough set theory to reduct attribution for mixed data information systems, we need to discretize the continuous data first. Otherwise, this will lead to the loss of some valuable information and the reduction results are not accurate. In order to overcome the mentioned problems, Lin has used the neighborhood relations instead of the traditional equivalence relation and has put forward the concept of neighborhood rough set. The neighborhood rough set can be defined as the neighborhood of an object, and it considers more about the correlation between samples and it doesn’t need to discretize. It has overcome the limitation of the classical rough set theory which can only deal with discrete data. Considering Lin’s neighborhood rough set theory and the incomplete situation existed in mixed data information systems, Baiting Zhao has put forward the tolerance relation based on neighborhood and related properties. This relation can well handle the missing value existed in mixed data information systems, but it only considers the similarities and dissimilarities between objects, it also affects the classification of the objects. Hengqiu Huang has put forward a tolerance relation of neighborhood connection degree based on the analysis of the above problems. The relation can artificially control the thickness of information granularity and have the advantage of scale, but it has only considered the influence of the identity degree and different degree to objects’classification without considering the influence of the opposite degree, so it can’t handle the mixture of incomplete information system with noise data well. Therefore, this paper carries on the further research, and the results of researches are highlighted as follows:(1)In this paper, we have analyzed the traditional rough set model and the neighborhood theory, in order to overcome the shortcoming of tolerance relation of neighborhood connection degree which haven’t considered the influence of opposite degree to objects’ classification, we introduced the opposite degree threshold, at the same time we have considered the influence of identity degree、different degree and opposite degree to objects’ classification and have put forward an improved tolerance relation of neighborhood connection degree, and we have used this relationship to determine classification of the mixed incomplete decision information system.(2)In order to overcome the disadvantage of the traditional rough set model to classify objects only considering full inclusion and exclusion, this paper has introduced the misclassification rate β to extend inclusion relation on the basis of improving tolerance relation of neighborhood connection degree and has put forward the rough set model of tolerance relation of neighborhood connection degree based on variable precision. In other word, this model allowed the existence of a certain degree of misclassification rate, and when we classified objects, it can be included to a certain extent. At the same time, we have defined the upper and lower approximation of the model and have given the character of this model. Finally we have illustrated the effectiveness and a relatively high classification accuracy of the model by example.(3)In order to solve the attributes reduction of mixed incomplete decision information system based on the improved neighborhood connection degree of tolerance relation, we have analysed the traditional binary discernibility matrix. Because of the inconsistency in decisions and the low efficiency of the delete method of the traditional binary discernibility matrix which can’t get the minimum attribute reduction either, this paper has put forward the improved methods of attribute reduction. Finally, we have verified the advantage of the improved attribute reduction method which has high attribute reduction efficiency and relatively optimal reduction results through practical examples and simulation experiments.