Method and system for automatic evaluation of feature relevance

摘要

The invention is concerned with a method or a system for automatic evaluation of feature relevance in a data (1001), especially such as stored datasets, online signals and / or data stream, characterized by:
a) specification of a feature parametrization (2001) indicative of feature relevance, continuous and differentiable with respect to feature values
b) computation of the relevance score, especially relevance parameters (1003) for each feature with respect to a target function of the inductive analysis by means of one-sided differentials, especially first- and/or higher-order differentials of the target function with respect to the parameters. With this it is possible to classify the data without having á priori information about the classification itself.