Research And Design Of A Data Mining System Platform Based On Plugin Architecture

With the great economic benefits produced by data mining technology,attention from academics and industries is comprehensively paid to it.Nowadays,several data mining systems have been developed and used practically.However,these systems are generally fit for experts from data mining and data analysis domain,capabilities and quantities of these integrated algorithms are fixed and the system components are not easy to extend,resulting into limitations in data mining system application. Therefore,to research integration techniques and methods for supporting considerable extensible data mining systems is theoretically important and practically valuable.Systems based on plugin architecture have the major feature of supporting extensibility,introducing this architecture approach into data mining system construction area will considerably meet the extensibility needs of the area.The current existing achievements in data mining system research area are generally analyzed and summarized by this dissertation,starting the study with data mining process model,the main features of the data mining system domain need are analyzed and a plugin-based architecture model is proposed to fulfill the extensibility requirements.The architecture model can support system extensibility horizontally and vertically and layered system analysis and design,leading to quite optimized and raised performance of the data mining system based on the architecture model.On the basis of the plugin architecture model,combining with object-orientation ideas and related design patterns,this dissertation carries out the structure design of plugin-based data mining system,describes the working mechanism of the system and conducts design on its major components such as core function packages,data object model,plugin algorithm configuration and the dataset and algorithm contract.In terms of the extensibility mechanism of these various system components,an adapter framework for multi-data source dealing and multi-algorithm integration is proposed by introduced the abstract factory pattern and the corresponding methods aiming at the key problems in the process of encapsulating,loading and managing these plugin algorithms.According to the previously finished system components analysis and design,the prototype system of the proposed plugin architecture based data mining system platform is implemented on the Java platform.The prototype system shows its basic interface structure and main components,and the prototype is applied to a real-world dataset to test and to further verify the proposed models and methods,achieving the preliminary goal of linkage between theoretical study and practical application.