Time Series Forecasting by Hybrid Soft Computing Model

Time series forecasting ranges from traditional to soft computing models. The traditional models include the statistics and econometrics models while soft computing models include neural networks (NN), fuzzy logic, and others. There are contradictory opinions in previous literatures on the superiority of the models. This paper extends a prior work on time series forecasting that used petrochemical product price data as a case study, in which it discussed NN method as soft computing model versus autoregressive integrated moving average (ARIMA) as traditional model for forecasting resin price. This paper proposes conducting the forecasting by using a hybrid neuro-fuzzy model that is adaptive neuro fuzzy inference system (ANFIS) and compares the result with the NN and ARIMA. Forecasting result shows that ANFIS model has a relative low error in terms of MAPE that is 1.06% and also a high directional accuracy which is 93%. Moreover, statistical test also shows that there is significant difference in ANFIS forecasting accuracy if compared to ARIMA and NN.