Research On Modeling Of Data Streams Mining Systems Based On Extended Predictive Model Markup Language

Data take the form of continuous data streams rather than traditional stored databases in a growing number of applications, including network traffic monitoring, network intrusion detection, sensor networks, fraudulent transaction detection, financial monitoring, etc. Researches on data streams mining have come to a new stage in terms of intelligent data analysis and gained great attention in recent years. People are interested in the potential rules in data streams such as association rules and decision rules. Different from traditional static data mining, distinct characteristics of streams data lead to many new challenges on data streams mining algorithms. However, researches on data streams mining run into an embarrassment because many algorithms about data streams mining have high efficiency but low usage. Compared with much work on developing algorithms for data streams mining, there is little attention paid on constructing data streams mining systems which use these algorithms in an efficient, rapid and intelligent way. From the standing point of modeling, this dissertation studies the construction of data streams mining systems.Firstly, this dissertation presents an extended predictive model markup language EPMML as a meta-modeling language. Considering the problems that current PMML lacks semantic description ability and holds large a number of language elements, an extended predictive model markup language EPMML is developed based on description logic. We design the description logic DL4PMML as the logical foundation of EPMML language, and then analyze the architecture and language elements of EPMML in details. Consequently, we prove the decidability and complexity of EPMML.Secondly, this dissertation presents the architecture of metadata about streams mining systems and defines the EPMML based data streams mining metadata. We analyze how to apply EPMML into the knowledge representation and knowledge reasoning respectively. An inconsistency checking framework for EPMML based data streams mining metadata is designed using its knowledge reasoning ability. We validate the well-formedness, well expressibility and reasoning efficiency of EPMML by concrete experiments.Thirdly, this dissertation presents the data modeling theory for data streams mining and expatiates approaches to apply EPMML into describing the data component of data streams mining systems. We present a formal data modeling theory for data streams mining, which interprets the rules extracting and knowledge discovery from streams data precisely. We propose the data structure model of data streams, analyze the intension and extension of concepts, expatiate the nature of the rules extracting and concept drifting upon data streams. Furthermore, we analyze and illustrate how to apply EPMML into meta-modeling data components of data streams mining systems by concrete experiments.Fourthly, this dissertation presents an algorithm management model for data streams mining and expatiate approaches to apply EPMML into describing the algorithms component of data streams mining systems. Combined with EPMML language, we design an algorithm management framework for data streams mining systems AMF-DSMS. In this framework, the data streams mining algorithms are designed as algorithm services. We analyze how to use EPMML to meta-modeling algorithm services and algorithm components. We validate the efficiency of AMF-DSMS by concrete experiments.Lastly, this dissertation designs the framework of the whole data streams mining system synthetically and expatiate the components and modules of the framework. The behavior semantics are designed based on the framework. Then we designed the modeling architecture for data streams mining systems. We analyze the functions of EPMML based data streams mining metadata for the system in details finally.