Study On Upper Bound Weak Ratio Rules

Data Mining is a rising research field involves multiple disciplines, it covers very wide application range, and nowadays, data mining is becoming a very active research subjects. Association rules mining is an important branch of data mining, which is used to describe the implicit relationship of the attributes in the transactional databases. As a natural attribute of database, there is wide foreground and huge application value for the quantitative attributes.In this paper, the association rules of quantitative attributes in database is dealt with, and a new association rules model named Upper Bound Weak Ratio Rules (UBWRR for short) is proposed. The scope of research includes model, properties, mining, reasoning and application. The main results are as follows:1. The relationship between Boolean association rules and UBWRR is researched on, and a conclusion is drawn, namely that as a special case of quantitative association rules, UBWRR is the generalization of Boolean association rules, and for each Upper Bound Weak Ratio Rule, there is a Boolean association rule acting as its support rule, which can partition upper bound weak ratio rules to different equivalence class.2. A depth-first algorithm for mining UBWRR, called UBBoundary, is proposed. Algorithm analysis and experiment results congruously show that UBWRR is able to accomplish the mission of mining UBWRRs. And furthermore,the Base of UBWRR is defined, which can derive a class of UBWRRs.3. UBWRR indicates uncertainty reasoning relationship between antecedent and consequent. Based on this relationship, two UBWRR uncertainty reasoning methods and their intuitive meaning are introduced. According to some known upper bound weak ratio rules and facts, the use of these two methods can reach corresponding conclusion which can be used to forecast the missing data and detect the outliers.4. By integrating the upper bound weak ratio rules and lower bound weak ratio rules, a new WRR method is obtained. Experiments show that the WRR method achieves better effect than LBWRR method in the application for reconstructing lost data, forecasting and outlier detection.