Research On The Related Problem Of Rule Extraction In Data Mining

The arrival of the information age will undoubtedly have a profound influence on our lives. It brings extensive quantity of data in which many important information and knowledge hide. How to get deep rules from data is the urgent problem which needs to be worked out. Rules represent the hidden nature of things, and can be used to predict or make decisions. Data mining is the new research field against such background, is the cross subjects of many areas, such as statistics, computer science, pattern recognition, artificial intelligence, machine learning, data base.Rule exaction is the traditional problem of data mining, and is the core problem, too. This article engaged in a serious study of the rule exaction problem.At first, the discretization of real value attributes is discussed. Whether a discretization process is reasonable determines the accuracy of expression and extraction for information. This paper discussed a series of Chi2 algorithm, proposed a new Integral-Chi2 algorithm with for discretization of real value attributes.Second, a method of extracting rules from information system used niche immunity algorithm based on sharing mechanism is proposed. The principle of sharing and competing resource in the sharing mechanism is consistent with the relation of complementing and contending among rules. Ordinary methods of confidence measure have some shortcomings. New conception of flexible confidence is introduced in this article, and the new measure method of flexible confidence is proposed.At last, neural network is a useful tool. Its nonlinear processing capability has very good performance. But the knowledge which the neural network learned is concealed in a large amount of connections. It is very hard to understand. A new method of extracting rules from information system by integrating rough sets and neural network is proposed in this article. The computational complexity is polynomial. And the method can be applied to discrete values.