Classificador simbÃlico baseado em regiÃes de tipo casca convexa

AUTOR(ES)

Simith Tupinambà D Oliveira JÃnior

DATA DE PUBLICAÇÃO

2005

RESUMO

On the recent advancement from the information sciences technologies, deferent techniques are introduced to synthesize, analyze and extract information knowledge stored in huge databases. The symbolic data analysis (SDA) is a domain in the area of automatic knowledge discovery (KDD), related with multivariate data analysis, pattern recognition, artificial intelligence and database. SDA aims to generalize methods of the exploratory data analysis, and the statistical techniques (factorial analysis, regression, classification, etc) to symbolic data. These new data are more complex than the classic ones. Therefore it contains internal variation and they are structuralized. This work introduces a classifier for quantitative values vectors based on convex hull regions. The approach central idea is to construct regions that describes and discriminates the class of observed examples. In the SDA literature related to symbolic classifiers based on regions, each class at the end of the learning step, is described by a convex hull (or set of regions) defined by the hyper-cube formed by the objects belonging to this class. This description is obtained by a symbolic operator (join) and a Mutual Neighborhood Graph. On the allocation step, the new observations are assigned, using different matching functions. On the proposed classifier, the description of each class is a region (or set of regions) in Rp defined by the convex hull formed by its objects. This new approach aims to reduce the over generalization that is produced when each class is described by a region (or set of regions) defined by the hyper-cube formed by the class objects and then, to improve the accuracy performance of the classifier. On the allocation step, each new observation is assigned to a class based on a dissimilarity function which compares the class description (a region or a set of regions) with a point in Rp.

ASSUNTO(S)

dados simbÃlicos symbolic data abordagem orientada à regiÃo convex hull region oriented approach anÃlise de dados simbÃlicos classificaÃÃo supervisionada symbolic data analysis casca convexa ciencia da computacao supervised classification

Documentos Relacionados

• Consumo simbÃlico no Orkut: associaÃÃo Ãs marcas de moda no ambiente virtual
• O poder simbÃlico e o ethos do jornalismo policial da Folha de Pernambuco
• A natureza do comÃrcio das regiÃes brasileiras no Mercosul
• Estimativa da SulfataÃÃo Total na Atmosfera em Diferentes RegiÃes do CearÃ
• Anatomia das regiÃes selar e perisselar: estudo por ressonÃncia magnÃtica