L0-Constrained Regression for Data Mining

Zhili Wu*, Chun-hung Li

*Corresponding author for this work

Research output: Chapter in book/report/conference proceedingConference proceedingpeer-review

7 Citations (Scopus)


L2 and L1 constrained regression methods, such as ridge regression and Lasso, have been generally known for their fitting ability. Recently, L0-constrained classifications have been used for feature selection and classifier construction. This paper proposes an L0 constrained regression method, which aims to minimize both the epsilon-insensitive fitting errors and L0 constraints on regression coefficients. Our L0-constrained regression can be efficiently approximated by successive linearization algorithm, and shows the favorable properties of selecting a compact set of fitting coefficients and tolerating small fitting errors. To make our L0 constrained regression generally applicable, the extension to nonlinear regression is also addressed in this paper.
Original languageEnglish
Title of host publicationAdvances in Knowledge Discovery and Data Mining - 11th Pacific-Asia Conference, PAKDD 2007, Proceedings
EditorsZhi-Hua Zhou, Hang Li, Qiang Yang
PublisherSpringer Verlag
Number of pages8
ISBN (Electronic)9783540717010
ISBN (Print)9783540717003, 3540717005
Publication statusPublished - 27 Apr 2007
Event11th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2007 - Nanjing, China
Duration: 22 May 200725 May 2007

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349


Conference11th Pacific-Asia Conference on Knowledge Discovery and Data Mining, PAKDD 2007

Scopus Subject Areas

  • Theoretical Computer Science
  • Computer Science(all)

User-Defined Keywords

  • Feature Selection
  • Ordinary Little Square
  • Nonlinear Regression
  • Support Vector Regression
  • Ridge Regression


Dive into the research topics of 'L0-Constrained Regression for Data Mining'. Together they form a unique fingerprint.

Cite this