Re-engineering the forecasting phase using traditional and soft computing methods

Abstract

The aim of the work is verifying the possibility of extrapolating information on demand trends, for a company specialized in the production of aluminium tins, using the data collected in previous periods. This study is mainly divided into three stages: (1) data pre-processing (data collection) stage, (2) adaptive network evaluating stage and (3) forecast and recall stage. At the stage of data collection, the data are divided into four categories: time serial data, macroeconomic data, downstream production demand data and industrial production data. The company analysed in this work usually carried out the prediction activities by means of expert judgement. In the case analyzed, four models were developed in order to predict the monthly number of tins: three traditional methods based on historical series and neural networks. Soft computing models were compared with traditional prediction models. Particularly the Holt-Winters forecasting method was tested developing a model that take into account seasonal phenomena.