SeleÃÃo de modelos de previsÃo baseada em informaÃÃes de desempenho

AUTOR(ES)

PatrÃcia Maforte dos Santos

DATA DE PUBLICAÇÃO

2006

RESUMO

A time series is defined as a sequence of observations, which are ordered in time. There are several real problems that can be represented as time series, such as the monthly water consumption, registered during a month; or the values of a financial application, measured during a week. The use of time series forecasting can occur in several areas, as the financial market, fraud detection, chemist industry, medicine, among others. There are several models that can be used to forecast a time series. Thus, the selection of the most apropriate model to use can be a difficult task, which depends on different factors as the adjustment of the candidate models parameters and the time series characteristics. We can find in the literature many approaches that are used for the selection of forecasting models. In this work, we used an Meta-Learning approach, initially developed to the selection of learning algorithms and adapted to the model selection problem. Differently from the most common solutions to model selection, the approach used indicates not only the best model which is applicable to the input problem, but it also provides a ranking of the candidate models based on performance criteria provided by user. The performance results obtained by the candidate models on past problems are used for the suggestion of models to new problems. Thus, the proposed solution is more informative, giving the user a better perception of the relation between the candidate models

ASSUNTO(S)

meta-learning sÃries temporais meta-aprendizado ciencia da computacao previsÃo forecasting time series

Documentos Relacionados

• Selection of common bean families based on grain type and yield, and QTLs informations
• RequalificaÃÃo urbana a partir de informaÃÃes de cadastros territoriais e sistemas de informaÃÃes geogrÃficas.
• Um Framework para extraÃÃo de informaÃÃes: uma abordagem baseada em XML
• ExtraÃÃo de informaÃÃes estruturais e de comunicaÃÃo de descriÃÃes em SystemC
• Modelagem 3D de faces humanas baseada em informaÃÃes extraÃdas de imagens bidimensionais nÃo calibradas.