Research Of Feature Extraction Algorithm Based On Rough Set

Knowledge induction is the core of intellectualized-decision-supporting system, that is, a process in which, the obtained information results in logical decision rules and useful knowledge by the data analysis and induction. Rough Set theory, as a new product of the intellectualized information disposal technique developed by a Polish scientist, Z.Pawlak, is a new method to analyze, induct, learn and discover the incomplete data. According to the method of RS, knowledge induction is to obtain all the minimum decision algorithms based on the condition-attributes and decision-attributes of given knowledge system. What is discussed in the paper is the problem---how to get the essential feature of things, or to find the simplest condition-attributes set and minimum decision algorithm.The method of Rough Set theory, better than the method of fuzzy set and probability statistic, is the effective tools to deal with inexact information. So it is very valuable in theory and practice to research how to achieve in the essence feature by the method of RS.The main content of the paper is as follows: Firstly, this paper introduces the knowledge representation system and decision tables and the concept of decision table reduction, then puts forward the algorithm for each phase of the reduction. Next, it expounds the design principle of feature extraction algorithm based on the RS and the method to put it into practice. Finally, the paper summarizes the main achievements and deficiencies, and puts forward some problems to be researched.This paper has done the rewarding try in the research of feature extraction algorithm based on RS. The main achievements contain: 1. Put forward the condition-attributes reduction algorithm of decision table and the algorithm to determine the core-value table;2. Bring forward the algorithm for the reduction of decision rules, by which we can obtain the set of all reduced rules;3. Determine the minimum decision algorithm: to improve the parameter k, judging the decision attributes depending on condition attributes and give a new parameter K, which can more embody the dependability;4.Develop a new distinguishing theory, which determines whether to reestablish the decision tree established by learning an original training set, which can distinguish whether the stability of the original decision tree will be changed or not, when a new example is added to the training set.