It decomposes at first a time series into a trend part and some detail parts using the EMD, and then divides all trend series into two sets: training sets and testing sets.