modeling and prediction of bread waste using time series models and artificial neural networks (ann)

this paper presents the application of multivariate time series model (ardl) to investigate factors affecting bread waste and to explore the relationships among shortrun, longrun and error correction coefficient and the independent variables over the period 1978-2006. results reveal that gross national product and urbanization have positive effects on bread waste in the long term, while the bread price and gini coefficient have negative effects on bread waste in short term. to predict the amount of bread waste, artificial neural network (ann) and ardl model were applied. comparison of the two models indicated that the ann-ardl multi-layer perceptron model (3 layers) with a hyperbolic tangent transfer function for the hidden layer and a delta-bar-delta learning algorithm, is the best model for forecasting the amount of bread waste. this amount will exceed 3.181 million tons in 2011. this implies that considering the wheat price in 2006, usd 1145 million will be removed from the national economic cycle. jel classification: c45, q13