Extension of multivariate regression trees to interval data. Application to elasticity load profiling (Q880915)

A regression tree growing algorithm is considered for data with multivariate interval-valued response and quantitative or nominal predictors. The splitting function is based on the Hausdorff metrics for intervals and least absolute deviations distance. The symbolic object principal component analysis procedure is recommended as a pre-processing algorithm. The developed technique is applied to the data on French customers electricity loading patterns.