DISCRIMINATIVE FEATURE SELECTION FOR DATA SEQUENCES

摘要

A discriminative feature selection method for selecting a set of features from a set of training data sequences (420) is described. The training data sequences (420) are generated by at least two data sources, and each data sequences consists of a sequence of data symbols taken as an alphabet. The method is performed by first building a suffix tree (300) from the training. The suffix tree (300) contains only suffixes of the data sequences having an empirical probability of occurrence greater than a first predetermined threshold, (430) from at least one of the sources. Next the suffix tree is pruned (310) on all suffixes for which there exists in the suffix tree (300) a shorter suffix having equivalent predictive capability, for all of the data sources.