Extended support vector interval regression networks for interval input-output data

In many applications, it is natural to use interval data because of uncertainty existence in the measurements, variability in defining terms (such as the temperature during a given day), description for extremely behavior (such as the maximum wind speed in a given area), etc. In order to handle such interval data, a novel approach, called the interval support vector interval regression networks (ISVIRNs), is proposed. The ISVIRNs is extended from our previous work, the support vector interval regression networks; SVIRNs. It is easy to find that SVIRNs can handle interval output data, but for input data, they must be crisp. In this study, the Hausdorff distance is employed as the distance measure of interval data and is incorporated into the kernel functions of SVIRNs to determine the initial structure of ISVIRNs. Because the proposed approach can provide a better initial structure for ISVIRNs, it can have a faster convergent speed. The experimental results with real data sets show the validity of the proposed ISVIRNs.