This paper presents a case study concerning the forecasting of monthly retail time series recorded by the US Census Bureau from 1992 to 2016. The modeling problem is tackled in two steps. First, original time series are de-trended by using a moving windows averaging approach. Subsequently, the residual time series are modeled by Non-linear Auto-Regressive (NAR) models, by using both Neuro-Fuzzy and Feed-Forward Neural Networks approaches. The goodness of the forecasting models, is objectively assessed by calculating the bias, the mae and the rmse errors. Finally, the model skill index is calculated considering the traditional persistent model as reference. Results show that there is a convenience in using the proposed approaches, compared to the reference one.
Titolo: | Forecasting Monthly Sales Retail Time Series: A Case Study |
Autori interni: | |
Data di pubblicazione: | 2017 |
Handle: | http://hdl.handle.net/20.500.11769/306674 |
Appare nelle tipologie: | 4.1 Contributo in Atti di convegno |