Research Of Multiple Classifiers System Based On Fusion Decision

Various classifiers can get various characteristic in the area of pattern recognition. But any of them can not get a satisfied result for all the applications. In order to improve the performance of single classifier, multi-classifiers fusion methods have been widely used. It complements various classifiers to improve the precision and stability effectively.This article not only talks about multiple classifiers system and gives some general ideas of the development and the theory framework, but also sum-up and organizes a certain amount of combinational function and base classifiers which were mentioned by many researchers. It analysis and compares these classic functions based on experimental points separately. The research results show that multiple classifiers system has been more effected by the research of the relationship between member of classifiers and the improvement of the whole procedure.This article brings up the design of the PCAB (Proof-Confidence based Attribute Bagging) based on the above analysis. PCAB optimizes the multiple classifiers system from classifies aggregate and fusion rule. The new fusion rule, betterment Bagging, assign weights to classifiers based on their performance on the training. Experiments with UCI datasets, compares with single classifier and other fusion algorithms, show that the performance of system is improved obvious, and the effectiveness of the novel algorithm.Final, the system of multiple classifiers fusion has been developed, using Visual C++. This system which has been developed has workability, can be a favorable assistant tool for the research of pattern recognition.