Quantitative Association Rules Algorithm Based On Fuzzy Sets

With the development of information, the problem of excessive information appears to us. People urgently need some new powerful, automatic and intelligent data analysis methods or tools to find out useful information and knowledge from vast data. Data mining is such a technology which can solve the problems effectively. Among all data mining tasks, association rules is one of the most important data mining tasks, which can discover all frequent models and strong association rules of data sets. At present, research on mining Boolean attributes with association rules has become more and more mature. However, research on mining quantitative attributes with association rules is still a hot topic.To solve the problem in mining quantitative data with association rules, it is an important method that converts quantitative attributes into Boolean attributes. We only need to map each value to Boolean item when the quantitative attributes have merely several values. Nevertheless, when dealing with those quantitative attributes which have wide value scope, we must divide the scope into several certain sectors at first, and then map every sector to Boolean item. However, sector division of quantitative data is not a simple process, and it can bring some problems. For example, if we divide the sectors into a smallest sector, it will cause some information lost and if we deal with some lopsided data, it will hardly to show the data distributing and so on. Therefore, the problem of association rules in data mining algorithm with quantitative data is a problem of how to divide the range of data attribute reasonably. Though many people have suggested solving this problem by fuzz theory, but the implementation of this algorithm is rare, which needs to be discussed and study.In this dissertation, firstly, we elaborated some issue about data mining and association rules, and then we discussed some fuzzy theory concept and characteristic in detail, such as fuzzy set, fuzzy similar matrixes and fuzzy equivalent matrixes. A new clustering algorithm with unfixed sector number, named FEM-TC, is presented which is based on fuzzy equivalent matrixes. We describe the general step to realize the algorithm and verify its correctness and effectiveness by using direct observation with typical data. After determining to use F statistic value as the judgment of clustering results, we choose its essential steps including data standardization and constructing fuzzy similar matrixes to make a comparison in many aspects and finally, we find out the best model. Besides the above work, another classification algorithm with fixed sector number, named FMI and its evaluation criteria are presented which based on ISODATA method. We stated its implementation specifically and we also use some typical data to test and analyze the correctness and effectiveness of the algorithm to get some requirements in dividing initial fuzzy matrix.After the research of two algorithms in sector division, the dissertation analyses the Apriori algorithm and make an improvement above it to solve the problem of how to scan database only once and attributes subset for generating association rules. Next, the dissertation provided a new method to calculate the support and confidence value based on degree of membership.In the last part of the dissertation, we applies these algorithms as mentioned above to a coal tax monitoring system to make some analysis, which can well verify the effectiveness and usability of the algorithms in some extend.