Forecasting Demand With Support Vector Regression Technique Incorporating Feature Selection in the Presence of Calendar Effect

Forecasting Demand With Support Vector Regression Technique Incorporating Feature Selection in the Presence of Calendar Effect

Malek Sarhani (ENSIAS, Mohammed V University, Morocco) and Abdellatif El Afia (ENSIAS, Mohammed V University, Morocco)
Copyright: © 2018 |Pages: 15
DOI: 10.4018/978-1-5225-5273-4.ch012

Abstract

Reliable prediction of future demand is needed to better manage and optimize supply chains. However, a difficulty of forecasting demand arises due to the fact that heterogeneous factors may affect it. Analyzing such data by using classical time series forecasting methods will fail to capture such dependency of factors. This chapter addresses these problems by examining the use of feature selection in forecasting using support vector regression while eliminating the calendar effect using X13-ARIMA-SEATS. The approach is investigated in three different case studies.
Chapter Preview
Top

Background

Modeling and Eliminating the Calendar Effect

When modeling a time series, the seasonal adjustment is necessary, this adjustment must also include the process for identifying in a time series the periodic components.

They can be from two types, the first one is any cyclical patterns that may evolve as the result of changes associated with the seasons, and the second one is related to factors which do not necessarily occur in the same month each year, this second effect is the calendar effect.

In our previous paper (Sarhani & El Afia, 2016), we have presented a review on the existing works on modeling the calendar effect. Also, on the paper of (Sarhani & El Afia, 2014a), we presented a comparison of the most used methods which are X-12-ARIMA and TRAMO-SEATS, we concluded the new approach X-13ARIMA-SEATS which combines the features of the two programs is the more adaptable for modeling the calendar effect.

Complete Chapter List

Search this Book:
Reset