Convex Optimization for Group Feature Selection in Networked Data

Daehan Won (), Hasan Manzour () and Wanpracha Chaovalitwongse ()
Additional contact information
Daehan Won: Systems Science and Industrial Engineering Department, Binghamton University, the State University of New York, New York, New York 13902
Hasan Manzour: Department of Industrial and Systems Engineering, University of Washington, Seattle, Washington 98195
Wanpracha Chaovalitwongse: Institute for Advanced Data Analytics, Department of Industrial Engineering, University of Arkansas, Fayetteville, Arkansas 72701

INFORMS Journal on Computing, 2020, vol. 32, issue 1, 182-198

Abstract: Feature selection is at the heart of machine learning, and it is effective at facilitating data interpretability and improving prediction performance by defying the curse of dimensionality. Group feature selection is often used to reveal relationships in structured data and provide better predictive power compared with the standard feature selection methods without consideration of the grouped structure. We study a group feature selection problem in networked data in which edge weights are considered as features, while each node in the network is regarded as a group feature. This problem is particularly useful in feature selection for neuroimaging data, where the data are high dimensional and the intrinsic networked structure among the features (i.e., connectivities between regions) in brain data has to be captured properly. We propose a mathematical model based on the support vector machines (SVM), which entails the ℓ 0 norm regularization to restrict the number of nodes (i.e., groups). To cope with the computational challenge of the ℓ 0 norm regularization, we develop a convex relaxation reformulation of the proposed model as a convex semiinfinite programming (SIP). We then introduce a new iterative algorithm that achieves an optimal solution for this convex SIP. Experimental results for synthetic and real brain network data sets show that our approach gives better predictive performance compared with the state-of-the-art group feature selection and the standard feature selection methods. Our technique additionally yields a sparse subnetwork solution that is easier to interpret than those obtained by other methods.

Keywords: semiinfinite programming; classification; multiple kernel learning; convex optimization; support vector machines (search for similar items in EconPapers)
Date: 2020
References: View references in EconPapers View complete reference list from CitEc
Citations: View citations in EconPapers (7)

Downloads: (external link)
https://doi.org/10.1287/ijoc.2018.0868 (application/pdf)

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:inm:orijoc:v:32:y:2020:i:1:p:182-198

Access Statistics for this article

More articles in INFORMS Journal on Computing from INFORMS Contact information at EDIRC.
Bibliographic data for series maintained by Chris Asher ().