Resumo:
Nonlinear time series forecasting is widely used in several areas to make good inferences
about the future and to support decisions. Many examples of nonlinear time series include
medical observations, financial recordings, and weather data. The accuracy of forecasts
is determined by considering how well a model performs on new data that were not used
when fitting the model and the monitoring of forecast errors is essential to ensure
forecasting accuracy. Therefore, this thesis presents a nonlinear time series prediction
methodology using Neural Networks and Tracking Signals method to detect bias and their
responsiveness to non-random changes in the time series. Datasets were generated to
simulate different nonlinear time series by changing the error of the series. The datasets
were predicted by Artificial Neural Network, Multilayer Perceptron, and the forecast
errors were monitored by Cumulative Sum Tracking Signals. Different from many studies
published in the area, the statistical methodology of Design of Experiments was applied
to evaluate the tracking signals based on Average Run Length. After, the methodology
was applied in data based on total oil and grease and it was compared with the application
of other traditional methodologies. The results showed that the proposed prediction
methodology is an effective way to detect bias in the process when an error is introduced
in the nonlinear time series because the mean and the standard deviation of the error have
a significant impact on the Average Run Length. This study contributes to a discussion
about time series prediction methodology since this new technique could be widely used
in several areas to improve forecast accuracy.
