Combination of Box-Jenkins and MLP/RNA models for forecasting

Abstract

This study aims to predict the values of the time series of UHT milk demand in a dairy industry by combining forecasting of ARIMA and MLP/RNA models and compare the results to the individual models, exemplifying the combined forecast for the production planning. Eight predictions combining techniques were used and, after the use of statistical techniques, the results obtained by fitting the ARIMA and MLP/RNA templates were compared with the results obtained in the proposed combinations. The results showed that the combination of SARIMA models (3,0,1)(1,1,0)12 and DMLP the inverse mean square method provided a performance in the forecast, for six months ahead, 66.5% higher than the individual models where the combination of forecasts provided a RMSE of 1.43 and MAPE of 2.16. The forecast for 12 months on, the performance of the combination was 56.5% higher compared to individual models, with RMSE of 2.86 and MAPE of 3.70%. In both cases, the combination of predictions showed superior results.