Development of data modification method for optimization of forecasting performance

Abstract

Dynamic nature of influencing parameters on market variations prevents decision makers to have a broad vision about possible future changes as an important factor in an organization survival. A precise forecast of both price and demand is a vital issue to illustrate market changes, and prosperity of plans and investments. The main purpose of this study is to develop a quantitative method, which encompasses human user cognition in order to modify timeseries, before being used as an input for forecast models. Some studies conclude ARIMA-ANN hybrid model as the best forecasting model in comparison with its individual models. However, this claim is rejected in some cases. It is a reason to check the performance of individual models in addition to hybrid model in new cases. Historical data are collected from two case studies in manufacturing and service industries. These data are modified by the developed method. Both original and modified data are implemented as inputs for ARIMA, artificial neural network (ANN), and ARIMA-ANN forecast models. The square errors (MSE) and mean absolute percentage error (MAPE). In both case erformance. In predictions