| Intelligent Decision Support Systems (IDSS), which is the combination of decision support systems and artificial intelligence, is a promising research area. Data mining is an important tool to improve its abilities of management decision support.Data mining is the process of discovering non-trivial, previously unknown, potentially useful and understandable information from large datasets. Data from the database of decision domain are often uncertain and incomplete, and resulting in difficulty of knowledge discovery from this kind of data. To solve above question, the dissertation suggests a method based on rough sets theory namely rough sets data analysis (RSDA) after detailedly researching the characters of uncertainty problems in the data mining process and carefully analyzing the advantages and disadvantages of several uncertainty theories. The dissertation studies the extending model and computational methods of rough sets, preprocessing of data, knowledge reduction and decision analysis etc. The details are as follows:1. The extending model of rough sets theory. Because of the shortcomings of the classical rough sets model such as the sensitive to noise often encountered in many real-world applications, the dissertation presents a variable precision and MD relation rough sets model from the perspective of rough membership function and micro-difference. Not only can the VP-MD model overcome the shortcomings of the classical model, but also is consistent with the statistics. This model can extend the application scopes of rough sets and enhance its adaptability.2. The computational methods of VP-MD model. The dissertation suggests a series of algorithms to compute the sets and numeric data based on the VP-MD model. To obtain the minimal reduction of attributes and the minimal reduction of values, the dissertation provides a CSBARK algorithm based on the context sensitivity (CS) of attributes and a minimal rules set algorithm based on value core. The CSBARK algorithm computes the CS of an attribute according to the weighted feature difference and value difference, then searching the minimal reduction of attributes using CS as heuristic information. The CS value not only represents the importance of a condition attribute, but also reveals its relationships with other attributes. The minimal rules set algorithm depends on the value core and integrates the logic relation in the reasoning. The dissertation also suggests an algorithm to compute the support and the confidence of a rule.3. This dissertation researches deeply the preprocessing of data from severalaspects:(1) Discretization of a set of continuous attributes. The paper changes the discretization problems to exploring the best cutting points based on genetic algorithmsfor discretizing the set of continuous attributes according to the basic concepts of rough set theory,then removing the redundant cuts and producing small discretizations which maintain the required capability of classification.(2) Concept generalization. A method is suggested in this paper that improves the concept-climbing algorithm by controlling the concept ascent degree with the use of the classification capability of an attribute. It can avoid overdoing the generalization.(3) Imputation of unknown values. To the incomplete information systems,the dissertation suggests a new method to impute the null values. The method is based on the extending model of rough sets and fully considers the consistence of data and dependency of attributes. This method can not only avoid deleting those entities including null values,but also avoid redundant extension of the incomplete information system.4. Using the minimal rules set,the dissertation proposes a decision classification method to class new objects according to the objects satisfying the rules set whether or not.5. To improve the decision ability of the GDSS,we will propose a new idea in this paper based on rough set and analytic hierarchy process to construct the judgment matrices us...
|
PDF Full Text Request