Knowledge Discovery Theory And Its Applications

With the existing of tremendous amount of data and the requirement of transforming the vast amount of data into useful information and knowledge, the knowledge discovery in databases (KDD), a process of identifying useful knowledge in data, has come to being a new subject with widespread application since 1990s'. However, the theoretical system of the KDD has not developed into a mature framework yet. And there is still a nontrivial gap between general principles of the KDD and domain-specific and effective KDD tools for particular applications. In this thesis, the KDD is studied, and methodologies in the KDD such as the fuzzy sets, the neural network and the rough sets are discussed. This thesis is organized as follows:First, the concepts and techniques of KDD are comprehensively introduced. As the general architecture of KDD is proposed, the functionalities and applications of KDD are discussed.Second, the rough sets, the neural network and the fuzzy sets are briefly discussed. Some proposed methods for KDD, such as the BP neural network based on rough sets, the fuzzy C-means, the partition index maximization and so on, are introduced and implemented in computer simulation. A new classification method based on the fuzzy set theory and the information gain analysis technique is presented and implemented in computer, which can achieve higher accuracy with less computation and without the input of the number of attributes for classification. A clustering approach based on the wavelet packet is proposed, which deals with the huge amount of spatial data efficiently, finds out arbitrary shapes, suppresses noises and outliers, and is insensitive with the order of input data. The efficiency of the proposed approach is shown by the experimental results on very large data sets.At last, the summary of total research is given, and prospect of KDD is also given.