Implementation of and experimental software for active selection of classification features.

Abstract

In some machine learning applications, obtaining data on the most predictive features is costly, but other features are readily available. Recently, first active learning approaches for this Actively Selecting Classification Features problem (ASCF) have been proposed. In this paper, we introduce a Python package that provides a framework for ASCF, including implementations of a supervised and an unsupervised selection approach, as well as a framework for performing experimental evaluations. This framework has been used in recent publications in the context of neuroimaging research on mental disorders, where its usefulness has been demonstrated in a simulated study design with MRI data.