Detecção de clusters irregulares através da não conectividade ponderada de grafos

• Ano de defesa: 2010
• Autor(a) principal: Spencer Barbosa da Silva
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; UFMG
• Palavras-chave em Português: Vigilância sindrômica; Compacidade geométrica; Clusters irregulares; Estatística Espacial Scan; Algoritmos multi-objetivo; Cluster espacial; Função de não-conectividade; Estatística
• Link de acesso: http://hdl.handle.net/1843/ICED-87BNBS

Irregularly shaped spatial clusters are di±cult to delineate. The most likely cluster often spreads in a great proportion of the map, playing a significant role in its geography. Methods employing the Kulldor's scan statistics, associated to penalization procedures were used to control the over freedom of the clusters. Penalization functions for cluster geometry and the level of the cluster's connectivity are recent proposals. The non-connectivity measurement is eficient when guiding the detection, however, it shows problems when interpreting the important role of the connections inside a possible cluster. This study presents a weighing strategy for the non-connectivity terms which maximizes its eficiency when detecting irregular clusters. Experiments using simulated data were undertaken in order to check the improvement when using the weighing version. The results show a significant improvement when compared to experiments which do not use the ponder version. This method can be very important in epidemiology studies and disease surveillance. Another important advantage of this proposal is the factthat it requires low computational time.