Study On Mining And Reasoning Of Weak Ratio Rules

Data Mining is a very active research frontier of intelligent information processing. It has successful applications in many areas. Mining association rules is one of the most attractive research subjects of data mining. In this paper, a special quantitative association rules called Weak Ratio Rules (simply WRR) is proposed. The study on WRR consists of five parts: model, property, mining, reasoning and application. The main research results are as follows:1. The properties of Goguen inclusion degree are discussed, and a new concept support degree is presented which can describe the degree of inclusion of two fuzzy subsets better than Goguen inclusion degree. The support degree of two fuzzy subsets is generalized to that of two [0, +∞]-valued fuzzy subsets. As a special case of the support degree of two [0, +∞]-valued fuzzy subsets, the support degree of two nonnegative real-valued functions on a finite set is used to describe weak ratio rules.2. Two algorithms, GenApriori and Boundary which functions are to find all elements of a down-set of direct product of finite finite-chains, are designed. GenApriori is a generalization of R.Agrawal's Apriori algorithm. GenApriori is a breadth-first search algorithm, but Boundary is a depth-first search algorithm. Algorithm analyses and experiments show that Boundary is more efficient than GenApriori in some cases and GenApriori is more efficient than Boundary in some other cases. The two algorithms are all used to mine quasi-maximal WRR.3. It is obtained that the WRR problem is a generalization of Boolean association rules problem and a specilization of quantitative association rules problem. It is proved that every WRR can induce a Boolean association rules as its support rule.4. Three WRR uncertainty reasoning methods and their intuitive meanings are introduced.5. WRR is applied to reconstructing lost data, forecasting and outlier detection. Experiments show that the application is successful.