Título: A Nonlinear Optimizated PSO-SVR Hybrid System for Time Series Forecasting with ARIMA
Autores: Emilly Alves, João Fausto de Oliveira, Manoel Marinho and Francisco Madeiro.
Resumo:
In the forecasting time series field, the combination of techniques to aid in predicting different patterns has been the subject of several studies. Hybrid models have been widely applied in this scenario, where the vast majority of series are composed of linear and nonlinear patterns. The Autoregressive Integrated Moving Average (ARIMA) presents satisfactory results in a linear pattern prediction but can not capture nonlinear ones. In dealing with nonlinear patterns, the SVR has shown promising results. In order to map both patterns, an optimized nonlinear combination model based on SVR and ARIMA is proposed. The main difference in comparison with other works is the use of an interactive PSO to increase the prediction performance. To the experimental setup, six well-known datasets of the literature are used. The performance was assessed by the metrics MSE, MAPE and, MAE. The results show the proposed system attains better outcomes when compared to the other tested techniques, for most of the used data.
Palavras-chave:
genetic algorithm, neural network architecture search, deep learning, evolutionary algorithms
Páginas: 8
Código DOI: 10.21528/CBIC2021-54
Artigo em pdf: CBIC_2021_paper_54.pdf
Arquivo BibTeX: CBIC_2021_54.bib