Dissertation > Excellent graduate degree dissertation topics show

Research on Incremental Mining Algorithm for Incomplete Data

Author: ZhangDeDong
Tutor: LiRenZuo
School: Lu Tung University
Course: Applied Computer Technology
Keywords: Incomplete data Rough Set Incremental technology Attribute core Attribute Reduction Rule acquisition
CLC: TP311.13
Type: Master's thesis
Year: 2010
Downloads: 44
Quote: 0
Read: Download Dissertation

Abstract

In real life, a dynamic database abound, which makes the incremental mining technology has become an important research field of data mining. Complete data on the incremental mining algorithm has achieved fruitful results, but incomplete data such research is still in the initial stage of the relevant research results at home and abroad is also unusual, carried out for the incomplete data growth mining algorithm has an important academic significance and broad prospect. The paper focuses on how the knowledge gained and incremental data fusion of the best match, in order to achieve efficient updating of knowledge issues. Mainly includes the following aspects: 1 Attribute nuclear incremental update: proposed the attribute core update algorithm based on discernable matrix. First define the approximate accuracy on incomplete data to determine the importance of an attribute metrics. And then based on the importance of the condition attributes, given the definition of an incomplete data attribute core. Then, based on the existing resolution matrix is ??constructed under an incomplete data improved discernibility matrix. Finally, improved discernibility matrix and attribute core definition, design attribute core of an incremental updating algorithm. Theoretical analysis and examples show that this method can quickly and efficiently incomplete decision table attribute core updates. 2 resolution function the paradigm transformation: to obtain a decision table based on discernibility matrix reduction process can be converted to distinguish function from CNF converted to disjunctive normal process, the efficiency of the performance of the algorithm for attribute reduction and rule extraction to off important. Based on the the artificial paradigm shift operation mechanism, take full advantage of the conjunctive operation and disjunction absorption rate, and with the queue structure, a distinguish-oriented paradigm of the function conversion algorithm. The algorithm is easy to understand, easy to achieve, simulation results show that the algorithm can efficiently paradigm conversion. 3 Attribute Reduction incremental update: proposed two algorithms. One is based on the generalization of the decision-making incremental algorithm. A way to get a first analysis of new data added generalization may cause a different variation of the decision-making, and then keep the same premise generalization decisions update strategies for different situations, and finally integrated the different update strategies designed incremental algorithm of attribute reduction. Another incremental algorithm based on discernable matrix. By analyzing the new object discernable matrix changes due to a variety of situations, distinguish different design matrix and resolution function Quickly update method, the incremental strike all attribute reduction algorithm. Theoretical analysis and examples show that the two algorithms can quickly update incomplete decision table attribute reduction. The 4 classification rules incremental obtain: First, by defining conjunct table for storage conjunct distinguish function updates the conversion rules for conjunctive entry table updates, and then the new object into the existing rule matching and conflict are two types, and finally can be connected according to the disjunction of conjunctive operation, given the conjunct table update strategies for different types of new objects, designed for incomplete decision table incremental rule updates algorithms. Batch obtained in increments of 5 classification rules: the traditional incremental Rules to add the object as the starting point for a problem faced by the algorithm, for the realization of the update of the rules, you need to access it again for each new object rule base. The rule base as the starting point, a decision rule bulk incremental updating algorithm. First of all, for all the new object the establishment of an equivalence class table, and then the original rule base equivalence class table for efficient matching last update rules different match types according to the new object. The algorithm applies to both complete data also applies to incomplete data, just visit twice a rule base can be achieved rules update. Theoretical analysis and comparison experiments on the UCI data show that the method is superior to traditional methods. In this paper, based on rough set theory system of incomplete data-oriented incremental mining problems, incomplete data on the incremental mining theory was extended, made a variety of effective mining algorithms, and the related theoretical proof and experimental validation incremental update knowledge on incomplete data offers a variety of practical and efficient solutions.

Related Dissertations

CLC: > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer software > Program design,software engineering > Programming > Database theory and systems
© 2012 www.DissertationTopic.Net Mobile