On Rough Set-based Knowledge Discovery And Decision-making Methods For Some Generalized Information Systems

With the continuous development of information science and technology,especially the rapid improvement of modern network technology and computer storage technology,the acquisition of data becomes more and more easy,resulting in the explosive growth of data scale.Meanwhile,the structure of data becomes more complex,then a series of generalized information systems are proposed to describe the large-scale complex data.How to quickly mine these large-scale and complex data,and take the corresponding decision analysis is of great significance in theoretical research and practical application.Rough set theory,a data-driven intelligent computing tool,which can conduct the knowledge discovery and decision analysis of data without prior knowledge,and it has obvious advantages for processing of uncertain data.Based on rough set theory,this dissertation studies the interval-valued ordered information system,multigranulation information system and hybrid-valued decision information system on knowledge discovery and decision analysis.The main research results and innovations of this dissertation are summarized as follows:(1)How to quickly update the approximations of interval-valued ordered information system when the attribute set vary dynamically.The dynamic compute approximations approaches are proposed associated with the deletion and insertion of attributes,and then the dynamic update approximations algorithms are designed for deleting and adding attributes,respectively.Furthermore,a series of experiments are conducted on several UCI datasets to verify the effectiveness of the designed algorithms,and the experimental achievements indicate that the dynamic methods significantly outperform the traditional approaches with a dramatic reduction in the computational efficiency.(2)How to conduct the double-quantitative decision analysis in multigranulation information system.Three pairs of double-quantitative multigranulation decision-theoretic rough set models are established by recombining the approximate operators of decisiontheoretic rough set and graded rough set models under the multigranulation framework.Then,the essential properties of these models are addressed,and the relationship between the double-quantitative multigranulation decision-theoretic rough set models and other models are discussed.Finally,a medical diagnosis case is conducted to exhibit the application of these models in decision analysis.(3)How to establish the decision-theoretic rough set model in hybrid-valued decision information system.Three types of different generalized hybrid distance measures are established,and then the neighborhood-based granulation mechanism is obtained by combining the Gaussian kernel function,meanwhile the thresholds setting rules in granulation process are also discussed.Moreover,the decision-theoretic rough set model is constructed by combining the Bayesian decision procedure in hybrid-valued decision information system,and then a practical case is demonstrated to exhibit the decision analysis process.(4)How to conduct the attribute reduction in hybrid-valued decision information system.Two attribute reduction methods associated with the relative positive region and minimum decision cost are studied,and then the corresponding heuristic attribute reduction algorithms are designed.Furthermore,different distance measures and reduction algorithms are adopted to obtain reduction based on several UCI data sets,and some comparisons are implemented in terms of reduction length and misclassification cost,respectively.In this dissertation,the knowledge discovery and decision analysis methods are studied for several generalized information systems based on rough set theory.The dynamic compute approximations approach is studied for the interval-valued ordered information system when attribute set vary dynamically,six kinds of double-quantitative decision-theoretic rough set models are established in multigranulation information system,the decision-theoretic rough set model and relative positive region and minimum decision cost-based attribute reduction approaches are studied in hybrid-valued decision information system.These findings expand and enrich the application of rough set theory in generalized information systems,provide a theoretical guidance and technical support for knowledge discovery in dynamic data environments and decision analysis of complex data circumstances.