The Improved Algorithm of ART2 in Data Mining
Liangun Li, Bin Zhang, Yuanyuan Che · 2009
Clustering analysis is an important research topic in data mining field, and it is one of main task of data mining. adaptive resonance theory (ART) neural network is an effective method to realize clustering. But the classical ART2 network has some shortcoming and insufficiency in data clustering application. The classical ART2 network must designate p alert parameters before the network training, the configuration of this parameter has a direct impact on the network clustering result. The classical ART2 uses the "winner takes all" competition rule, in general only considers winning neuron information, but neglects other useful neuron information in the output layer. The classical ART2 network output is essentially one-dimension structure, is unable to manifest the whole relation to entire input mode space. By improving the structure of ART2, ample considering amplitude information of mining object, which can decrease the requirement of vigilance parameter and earn cluster result with administrative level structure .The validity of this improve is verified by experimental result.