Research Into MAS-based Data Mining Algorithm Selection Mechanism

Nowadays, Data Mining technology gets more and more widely used in many areas, thus it is required much more to the research on data mining algorithms and proposed higher requirements to it. As a result, new data mining algorithms and technology emerges endlessly. Generally speaking, there is no one universally applicable data mining algorithm. So it is very important to select suitable data mining algorithms according to their own characteristics and scope of application, which directly affects mining efficiency and the quality of knowledge discovered. In short, it's really important and hard to choose a more efficiency and suitable data mining algorithm for a specific application in data mining field.This thesis has researched the automatic selection problem of data mining algorithm, put forward an intelligent selection model about data mining algorithm, and realized intelligent selection towards specific application object. In the process of model realization, AS technology is applied to reduce manual intervention and fluctuate data mining method dynamically considering the flexibility of data mining algorithm selection, As well as knowledge base and inference engine, which makes sure the intelligence of mining algorithm selection mechanism and the extensibility of selection method..In the designing of expert knowledge base, we has studied expert knowledge's representation, storage and utilization, all of which is based on relational database. Meanwhile, we has managed to organize a variety of related knowledge about data mining algorithm selection through matured management technology of relational database, determined the structure of knowledge base of data mining algorithm selection through production rules, realized automatic selection of data mining method through inference engine.In order to judge the effect of data mining algorithm selection, we has also studied evaluation criteria about data mining algorithm. According to the evaluation of data analysis and encapsulation methods, which considers the efficiency of all the positive features and negative features in data mining algorithms, to evaluate the algorithm's adaptability and efficiency,as well as determining the selected algorithm whether is the optimal algorithm.To prove the feasibility,we take"early warning model in oil development fracturing measures adjustment"as an instance and realize mining algorithm selection and complete early warning's mining.