Feature Selection for Time Series Forecasting: A Case Study

The integration of feature selection techniques within the modeling process of a time series forecaster can improve dealing with some usual important problems in this type of tasks, such as noise reduction, the curse of dimensionality and reducing the complexity of both the problem and the solution. In this paper we show how a convenient combination of feature selection procedures with soft computing techniques can be used to solve satisfactorily a real world problem. The problem is a rather hard one and consists of forecasting the amount of incoming calls for an emergency call center, so that the center managers can make a better resource planning.