With the great progress of microelectronics and other relating information technologies, together with the still broadening applications of computers in a vast range of businesses and industries, large databases containing mixed-mode data are becoming quite commonplace. Today, large databases contain various modes of collected data related to different components of a complex real world system. Their use is not necessarily confined to classifications. Many of them may not have clearly-defined class labelsor even any explicit class information at all. Indeed, there are many different reasons todetermine or discover all patterns, to achieve any comprehensive analysis and understanding of the information within the data spaces. In the past, data mining or pattern discovery has by and large been developed fundamentally for categorical databases. All of the classification rules have been found from pre-labeled data samples. When mixed-mode data are processed, engineers naturally work on the class-dependence relationship to discretize the real data. Where class information is lacking, there is no suitable way to discover patterns within these mixed-type databases. Consequently, most important pattern analysis jobs -such as pattern clustering, or even pattern summarization -being developed for categorical data will not be easily applied to a mixed-mode database. To break this impasse is theobjective of this thesis. We have attempted to develop some pattern discovery methods for mixed-mode databases where classes or features are unavailable. Analyzing these mixed-modes of databases and providing researchers with helpful knowledge is a challenging task. Developing new ways to turn the raw data into useful knowledge is now a long-term challenge in the data mining community
展开▼
