Research On Automatic Selection Methods Of Data Mining Models Based On Mas

Every sector accumulates massive amounts of business data, and it is imperative to convert those into useful information and knowledge, thus data mining has aroused enormous attention in the information industry. The choice and design of data mining models are the main links, and also key points to handle application problems effectively, when data mining technique is applied to solve kinds of practical problems. Traditional data mining model design relies on professional knowledge of modelers, and mining models are made via repetitive data exploration and algorithm testing on the basis of service features in application fields, so the working efficiency and modeling accuracy are reduced greatly. With the emergence of new techniques, model designers may ignore some important mining methods and algorithms which are helpful to knowledge discovery. Aiming at such problems as poor efficiency of data mining modeling artificially and the difficulty of knowledge reuse, this paper investigates application features, technical characteristics, and service features of data mining, explores the automatic modeling methods of data mining models, and designs the evaluation system of data mining models. Also, the automatic selection framework of data mining models based on MAS is built, combined with MAS technique on the basis of data mining automatic modeling methods, which is applied in oil field exploration area successfully.First, the basic meanings of the feature, framework, objective, activity, method and entity are clarified, and the data mining model selection general technology is developed via modeling concepts concerned with data mining model automatic selection methods. The concrete abstract and definition of data characteristics, service features and data mining technique characteristics are completed, and the characteristic system is established in the form of symbols. The automatic selection framework of data mining models is investigated combined with object analysis, activity analysis and method design based on node pattern. Making the data mining model selection and design as the overall goal, the mining behavior can be abstracted as five basic activities including data pre-processing, initial model design, model adjustment, model evaluation and knowledge representation. The data mining model selection framework specifies the objectives and activities during every stage of data mining model automatic design, and the specific logic relationship between mining businesses and mining technique characteristics. The framework design is based on the basic concepts such as objectives, activities and methods. The required expectations, responses and measures in different scenes during the mining process, and the model design activities of mining model selection at different stages and levels are expressed by node. The solving methods based on DMMS_F (Data Mining Model Selection Based on Features) and DMMS_E (Data Mining Model Selection Based on Experience) are designed in solving the feasible data mining model collections. Next, the data mining model evaluation system structure is designed, the comparatively applicable evaluation method of mining models is chosen from the feasible data mining model collections, the mining model evaluation objective is modeled, and the model evaluation objective in the mining model automatic selection mechanisms is described specifically. The comprehensive evaluation method of data mining models is explored, including the design of evaluation frameworks and evaluation factors. Taking into account the effect of subjective factors and objective factors on mining model evaluation, the evaluation framework based on the hierarchy is designed. The adjustable design method is adopted for the hierarchy position and weight of evaluation factors, the data mining model evaluation system is given, and the evaluation method of mining model quality is designed.Subsequently, the MAS technique is introduced to study the data mining model automatic selection methods, and the DMMAS (Data Mining Model Auto Selection) model framework based on MAS is set up. The concept and design of Agent cluster are proposed, and the collaboration and interaction of Agent cluster in the process of mining model selection design are implemented via the introduction of diplomatic role, management role and labor role. Afterwards, the support environment of data mining model selection is researched, the reasoning and running of data mining model design projects are separated, and the design platform used to finish feasible effective data mining project selection and configuration is built on the basis of rational knowledge organization. In accomplishing Agent technique, the design of Agent architecture and Agent collaborative models are explored. The Agent dynamic management platform is designed based on logic-ring organization structures. Also, the Agnet dynamic management is completed, mainly including the ring organization structures of Agent dynamic management platform and Agent management methods. The DMMAS system design based on MAS is analyzed from the perspective of system development, on the basis of model research.Finally, the application of DMMAS based on MAS in the oil filed exploration area is discussed. The oil well logging lithology identification mining model selection framework and the well selection system for fracturing measures are designed, and its quality is evaluated from the viewpoint of the operation and application. In designing mining models, the characteristic compliance system of data mining model general selection modeling is applied to give specific model selection flow and results, and then the service definition, automatic data selection, mining model design, model comparison and other functions of data mining automatic selection are accomplished from the perspective of development, thus instantiating the DMMAS model based on MAS.