GFEL: Generalized Feature Embedding Learning Using Weighted Instance Matching

Abstract

Feature embedding is an emerging research area which intends to transform features from the original space into a new space to support effective learning. Many feature embedding algorithms exist, but they are often designed to handle a single type of feature, or users have to clearly separate features into different feature views and supply such information for feature embedding learning. In this paper, we propose a generalized feature embedding learning algorithm, GFEL, which learns feature embedding from any type of data or data with mixed feature types. GFEL is an eigendecomposition based approach, which calculates weighted instance matching in the original feature space, and then uses an eigenvector decomposition to convert the proximity matrix into a low-dimensional space. The learned numerical embedding features, which blend the original features, can be directly used to represent instances for effective learning. Our experiments and comparisons on 28 datasets, including categorical, numerical, and ordinal features, demonstrate that embedding features learned from GFEL can effectively represent the original instances for clustering and classification tasks.