Taylor (2007) developed an exponentially weighted quantile
regression method, which generates interval forecasts from
quantile predictions. He also recommended the application of
interval forecasting method when the time series are highly volatile and skewed due to
extreme values. His method provides better results when compared to traditional methods. Chen
and Ou (2009) developed a model which integrates gray relation analysis and multi-layer
functional link network to forecast sales of perishable food in a convenience store. They proved
this model reduces more percentage of forecast error than other statistical time series (moving
average, autoregressive integrated moving average (ARIMA), generalized autoregressive conditional
hetero- skedasticity (GARCH)) and artificial neural networks (ANN) (BPNN, Generalized BPNN) models.
Hasin et al. (2011) suggested a fuzzy ANN approach to forecast sales of selected products
(including perishable foods) in a retail chain in Bangladesh. They also proved that the
forecasting performance of fuzzy ANN is better than holt- winter's model in terms of mean
absolute percentage error (MAPE). Lee et al. (2012) used a sales forecasting model for con-
venience stores using BPNN and compared the results with mov- ing average and logistic regression.
Shukla and Jharkharia (2013) applied ARIMA model to forecast the demand of vegetable on daily basis
in an Indian wholesale vegetable market with a MAPE in the range of 30%. Žliobaitė et al. (2012)
presented a two level switch model and studied a case of food wholesaler Sligro Food Group N.
V. This sales prediction approach divides the time series sales into predictable and random, then
uses intelligent predictor for pre-
dictable and moving average for random.