Referência Completa


Título: MLEM on cross section data.
Autor : Denise Beatriz Teixeira Pinto do Areal Ferrari
Curso : Engenharia Mecânica-Aeronáutica
Orientadores : Armando Zeferino Milioni
  Cairo Lúcio Nascimento Júnior
Ano de Publicação : 2003
Assuntos : Modelos matemáticos
t Indicadores socioeconômicos
t Produto interno bruto
t Análise de regressão
t Redes neurais
t Mapas auto-organizáveis (redes neurais)
t Macroeconomia
t Pesquisa operacional
t Matemática
Resumo : The Gross Domestic Product (GDP) is a macroeconomic indicator that synthesizes the final result of the productive activities of a country. However, the municipal GDP estimation is a very complicated procedure and often leads to imprecise results. The objective of this work is to investigate whether the very arduous task of estimating the municipal Gross Domestic Product (MGDP) can be replaced by the MLEM (Mixture of Local Expert Model) approach, in particular case. The case study considered is modeling the Gross Domestic Product on cross section data for the state of Pará, using relative easily available macroeconomic and demographic municipal variables to as the MLEM independent ones. In this study, the models used to compose the MLEM were the OLS (Ordinary Least Squares) and MSAE (Minimum Sum of Absolute Errors) regressions and a Back-Propagation Neural Network. Initially, all models are applied to a set containing all data points, and tests are carried out in order to determine the best-fitted model, called global expert. The global data set is then clustered using a Kohonen Neural Network and the three modeling techniques are applied to each cluster. Tests are performed, so that it is possible to select the local experts. A validation test is conducted using a data set comprised of 9 municipalities from the state of Maranhão, by applying the local experts previously calibrated.
Texto na íntegra : [Visualizar]