Study And Application Of Knowledge Reduction Algorithms Based On Indiscernibility Relation And Dominance Relation

Recently, with the rapid development of computer science and networkcommunication, the computerizing gains popularization in all aspects of our livesand thus comes large scale data sources. Taking into consideration the cost of timeand space, traditional mathematical analysis based on statistics can hardly deal withthe large amount of data which increase in amazing speed. Therefore, KnowledgeDiscovery in Databases which can intellectually and automatically abstract thevaluable knowledge from the large amount of data and its core technology— datamining deserve a further and comprehensive study.The rough set theory is a new kind of data mining methods which deals with theinformation under uncertainty. It can be used to find out the rules behind the datawithoutany addtionalknowledge. Reductionalgorithms are the key point of thismethod. In this paper, after discussingclassical rough set theory based onIndiscernibility Relation and its reductionalgorithms, the extended model of roughset theory for multiple criteria decision analysis based on dominance relation isstudied,including a reduction algorithm related to this model .It turns out that thereductionalgorithm leaves something to be desired.To show this,a counter exampleis thus presented to make clear the faults ofthis algorithm. To solve this problem,anew definition of dominance discernibility matrix is given by taking advantage ofthe unique characteristic of the extended model. The corresponding core andreductionalgorithms are accordingly presented and theoretically proved correct.Decision tree, another commonly used data mining method, is studied in this thesis.And MPID, an advanced decision tree algorithm based on merging branches is putforward after studying the classical ID3 algorithm and its optimized algorithmMID3. iiEmploying the theoretical research mentioned above, our research group implementsa data mining system called Hsminer using the web services technology as theplatform. An application instance of Hsminer is attached to the end of this thesis. Inthis instance, we use the rough set theory based on Indiscernability Relation to minethe rules in the remote sensing images of Fujian province.