A Comparative of a New Hybrid Based on Neural Networks and SARIMA Models for Time Series Forecasting

Abstract

AbstractIn this paper, we present a new hybrid forecasting model using decomposition method with SARIMA model and Artificial Neural Network (ANN). The proposed model has combined linear and non-linear models such as decomposition method with SARIMA model and ANN. The new hybrid model is compared to RBF, ANN, SARIMA, decomposition method with SARIMA models and RBF. We applied the new hybrid forecasting model to real monthly data sets such that the electricity consumption in the provincial area of Thailand and the SET index. The result shows that the new hybrid forecasting using decomposition method with SARIMA model and ANN can perform well. The best hybrid model has reduced average error rate for 3 months, 12 months and 24 months lead time forecasting of 93.3763%, 81.3731% and 74.3962%, respectively. In addition, the new hybrid forecasting model between decomposition method with SARIMA models and ANN has the lowest average MAPE of 0.1419% for 3 months, 0.4472% for 12 months and 1.7600% for 24 months lead time forecasting, respectively. The best forecasting model has been checked by using residual analysis. We conclude that the combined model is an effective way to improve more accurate forecasting than a single forecasting method.KeywordsDecomposition methodSARIMAArtificial neural networkRadial basis functionHybrid model